初始化项目，由ModelHub XC社区提供模型
Model: ayh015/myLightningOPD Source: Original Platform
This commit is contained in:
ModelHub XC
42
.gitattributes
vendored
Normal file
42
.gitattributes
vendored
Normal file
@@ -0,0 +1,42 @@
|
||||
*.7z filter=lfs diff=lfs merge=lfs -text
|
||||
*.arrow filter=lfs diff=lfs merge=lfs -text
|
||||
*.bin filter=lfs diff=lfs merge=lfs -text
|
||||
*.bz2 filter=lfs diff=lfs merge=lfs -text
|
||||
*.ckpt filter=lfs diff=lfs merge=lfs -text
|
||||
*.ftz filter=lfs diff=lfs merge=lfs -text
|
||||
*.gz filter=lfs diff=lfs merge=lfs -text
|
||||
*.h5 filter=lfs diff=lfs merge=lfs -text
|
||||
*.joblib filter=lfs diff=lfs merge=lfs -text
|
||||
*.lfs.* filter=lfs diff=lfs merge=lfs -text
|
||||
*.mlmodel filter=lfs diff=lfs merge=lfs -text
|
||||
*.model filter=lfs diff=lfs merge=lfs -text
|
||||
*.msgpack filter=lfs diff=lfs merge=lfs -text
|
||||
*.npy filter=lfs diff=lfs merge=lfs -text
|
||||
*.npz filter=lfs diff=lfs merge=lfs -text
|
||||
*.onnx filter=lfs diff=lfs merge=lfs -text
|
||||
*.ot filter=lfs diff=lfs merge=lfs -text
|
||||
*.parquet filter=lfs diff=lfs merge=lfs -text
|
||||
*.pb filter=lfs diff=lfs merge=lfs -text
|
||||
*.pickle filter=lfs diff=lfs merge=lfs -text
|
||||
*.pkl filter=lfs diff=lfs merge=lfs -text
|
||||
*.pt filter=lfs diff=lfs merge=lfs -text
|
||||
*.pth filter=lfs diff=lfs merge=lfs -text
|
||||
*.rar filter=lfs diff=lfs merge=lfs -text
|
||||
*.safetensors filter=lfs diff=lfs merge=lfs -text
|
||||
saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
||||
*.tar.* filter=lfs diff=lfs merge=lfs -text
|
||||
*.tar filter=lfs diff=lfs merge=lfs -text
|
||||
*.tflite filter=lfs diff=lfs merge=lfs -text
|
||||
*.tgz filter=lfs diff=lfs merge=lfs -text
|
||||
*.wasm filter=lfs diff=lfs merge=lfs -text
|
||||
*.xz filter=lfs diff=lfs merge=lfs -text
|
||||
*.zip filter=lfs diff=lfs merge=lfs -text
|
||||
*.zst filter=lfs diff=lfs merge=lfs -text
|
||||
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
||||
assets/intuitive.png filter=lfs diff=lfs merge=lfs -text
|
||||
assets/main-results.png filter=lfs diff=lfs merge=lfs -text
|
||||
assets/overview.png filter=lfs diff=lfs merge=lfs -text
|
||||
configs/sft/data/sft_data/openthoughts3_300k_qwen3-8b.jsonl filter=lfs diff=lfs merge=lfs -text
|
||||
data/prompts/dapo-math-17k/dapo-math-17k.jsonl filter=lfs diff=lfs merge=lfs -text
|
||||
data/prompts/openthoughts3_300k.jsonl filter=lfs diff=lfs merge=lfs -text
|
||||
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
||||
61
README.md
Normal file
61
README.md
Normal file
@@ -0,0 +1,61 @@
|
||||
---
|
||||
library_name: transformers
|
||||
license: other
|
||||
base_model: model_weights/qwen3-4b-base
|
||||
tags:
|
||||
- llama-factory
|
||||
- full
|
||||
- generated_from_trainer
|
||||
model-index:
|
||||
- name: qwen3-4b-base-sft-qwen3-8b
|
||||
results: []
|
||||
---
|
||||
|
||||
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
||||
should probably proofread and complete it, then remove this comment. -->
|
||||
|
||||
# qwen3-4b-base-sft-qwen3-8b
|
||||
|
||||
This model is a fine-tuned version of [model_weights/qwen3-4b-base](https://huggingface.co/model_weights/qwen3-4b-base) on the openthoughts3_300k_qwen3-8b dataset.
|
||||
|
||||
## Model description
|
||||
|
||||
More information needed
|
||||
|
||||
## Intended uses & limitations
|
||||
|
||||
More information needed
|
||||
|
||||
## Training and evaluation data
|
||||
|
||||
More information needed
|
||||
|
||||
## Training procedure
|
||||
|
||||
### Training hyperparameters
|
||||
|
||||
The following hyperparameters were used during training:
|
||||
- learning_rate: 8e-05
|
||||
- train_batch_size: 4
|
||||
- eval_batch_size: 8
|
||||
- seed: 42
|
||||
- distributed_type: multi-GPU
|
||||
- num_devices: 4
|
||||
- gradient_accumulation_steps: 2
|
||||
- total_train_batch_size: 32
|
||||
- total_eval_batch_size: 32
|
||||
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
||||
- lr_scheduler_type: cosine
|
||||
- lr_scheduler_warmup_ratio: 0.1
|
||||
- training_steps: 3000
|
||||
|
||||
### Training results
|
||||
|
||||
|
||||
|
||||
### Framework versions
|
||||
|
||||
- Transformers 4.52.4
|
||||
- Pytorch 2.12.0+cu130
|
||||
- Datasets 3.6.0
|
||||
- Tokenizers 0.21.1
|
||||
64
THIRD_PARTY_NOTICES.md
Normal file
64
THIRD_PARTY_NOTICES.md
Normal file
@@ -0,0 +1,64 @@
|
||||
# Third-Party Notices
|
||||
|
||||
This project includes or depends on third-party open source software. The following notices are provided in compliance with the respective licenses.
|
||||
|
||||
---
|
||||
|
||||
## Third-Party Code Included in This Repository
|
||||
|
||||
### verl (Bytedance)
|
||||
|
||||
- **File**: `slime/utils/seqlen_balancing.py`
|
||||
- **Source**: https://github.com/volcengine/verl/blob/468adf22c43b744348051fccd7a5d830c6c3c36a/verl/utils/seqlen_balancing.py
|
||||
- **Copyright**: Copyright 2024 Bytedance Ltd. and/or its affiliates
|
||||
- **License**: Apache License 2.0
|
||||
|
||||
### lm-evaluation-harness (EleutherAI / HuggingFace)
|
||||
|
||||
- **File**: `slime/rollout/rm_hub/math_dapo_utils.py`
|
||||
- **Source**: https://github.com/EleutherAI/lm-evaluation-harness/blob/main/lm_eval/tasks/hendrycks_math/utils.py
|
||||
- **Copyright**: Copyright 2024 Bytedance Ltd. and/or its affiliates; Copyright 2022 EleutherAI and the HuggingFace Inc. team. All rights reserved.
|
||||
- **License**: Apache License 2.0
|
||||
|
||||
---
|
||||
|
||||
## Third-Party Dependencies
|
||||
|
||||
This project depends on the following third-party packages, installed via `pip`. These are not distributed with this repository.
|
||||
|
||||
| Package | License |
|
||||
|---------|---------|
|
||||
| [PyTorch](https://github.com/pytorch/pytorch) | BSD-3-Clause |
|
||||
| [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) | BSD-3-Clause |
|
||||
| [Megatron-Bridge](https://github.com/fzyzcjy/Megatron-Bridge) | Apache-2.0 |
|
||||
| [SGLang](https://github.com/sgl-project/sglang) | Apache-2.0 |
|
||||
| [sglang-router](https://github.com/sgl-project/sglang) | Apache-2.0 |
|
||||
| [Ray](https://github.com/ray-project/ray) | Apache-2.0 |
|
||||
| [Transformers](https://github.com/huggingface/transformers) | Apache-2.0 |
|
||||
| [vLLM](https://github.com/vllm-project/vllm) | Apache-2.0 |
|
||||
| [LlamaFactory](https://github.com/hiyouga/LLaMA-Factory) | Apache-2.0 |
|
||||
| [Datasets](https://github.com/huggingface/datasets) | Apache-2.0 |
|
||||
| [Accelerate](https://github.com/huggingface/accelerate) | Apache-2.0 |
|
||||
| [NVIDIA ModelOpt](https://github.com/NVIDIA/TensorRT-Model-Optimizer) | Apache-2.0 |
|
||||
| [Weights & Biases](https://github.com/wandb/wandb) | MIT |
|
||||
| [TensorBoard](https://github.com/tensorflow/tensorboard) | Apache-2.0 |
|
||||
| [Liger Kernel](https://github.com/linkedin/Liger-Kernel) | BSD-2-Clause |
|
||||
| [ring-flash-attn](https://github.com/zhuzilin/ring-flash-attention) | Apache-2.0 |
|
||||
| [OmegaConf](https://github.com/omry/omegaconf) | BSD-3-Clause |
|
||||
| [PyYAML](https://github.com/yaml/pyyaml) | MIT |
|
||||
| [httpx](https://github.com/encode/httpx) | BSD-3-Clause |
|
||||
| [Pillow](https://github.com/python-pillow/Pillow) | MIT-CMU |
|
||||
| [pylatexenc](https://github.com/phfaist/pylatexenc) | MIT |
|
||||
| [pandas](https://github.com/pandas-dev/pandas) | BSD-3-Clause |
|
||||
| [PyArrow](https://github.com/apache/arrow) | Apache-2.0 |
|
||||
| [tqdm](https://github.com/tqdm/tqdm) | MPL-2.0 / MIT |
|
||||
| [blobfile](https://github.com/christopher-hesse/blobfile) | Unlicense |
|
||||
| [memray](https://github.com/bloomberg/memray) | Apache-2.0 |
|
||||
| [MCP](https://github.com/modelcontextprotocol/python-sdk) | MIT |
|
||||
| [Triton](https://github.com/triton-lang/triton) | MIT |
|
||||
|
||||
---
|
||||
|
||||
## Apache License 2.0 (Full Text)
|
||||
|
||||
The full text of the Apache License 2.0 is available in the [LICENSE](./LICENSE) file included with this repository.
|
||||
28
added_tokens.json
Normal file
28
added_tokens.json
Normal file
@@ -0,0 +1,28 @@
|
||||
{
|
||||
"</think>": 151668,
|
||||
"</tool_call>": 151658,
|
||||
"</tool_response>": 151666,
|
||||
"<think>": 151667,
|
||||
"<tool_call>": 151657,
|
||||
"<tool_response>": 151665,
|
||||
"<|box_end|>": 151649,
|
||||
"<|box_start|>": 151648,
|
||||
"<|endoftext|>": 151643,
|
||||
"<|file_sep|>": 151664,
|
||||
"<|fim_middle|>": 151660,
|
||||
"<|fim_pad|>": 151662,
|
||||
"<|fim_prefix|>": 151659,
|
||||
"<|fim_suffix|>": 151661,
|
||||
"<|im_end|>": 151645,
|
||||
"<|im_start|>": 151644,
|
||||
"<|image_pad|>": 151655,
|
||||
"<|object_ref_end|>": 151647,
|
||||
"<|object_ref_start|>": 151646,
|
||||
"<|quad_end|>": 151651,
|
||||
"<|quad_start|>": 151650,
|
||||
"<|repo_name|>": 151663,
|
||||
"<|video_pad|>": 151656,
|
||||
"<|vision_end|>": 151653,
|
||||
"<|vision_pad|>": 151654,
|
||||
"<|vision_start|>": 151652
|
||||
}
|
||||
8
all_results.json
Normal file
8
all_results.json
Normal file
@@ -0,0 +1,8 @@
|
||||
{
|
||||
"epoch": 0.34443168771526983,
|
||||
"total_flos": 3.429011909561549e+19,
|
||||
"train_loss": 0.27564545996983847,
|
||||
"train_runtime": 45365.0148,
|
||||
"train_samples_per_second": 2.116,
|
||||
"train_steps_per_second": 0.066
|
||||
}
|
||||
BIN
assets/cost.png
Normal file
BIN
assets/cost.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 54 KiB |
3
assets/intuitive.png
Normal file
3
assets/intuitive.png
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:985122a8b8e98092f7b4eb96fa5777c2de2d1238f10a0db4fd9821618797775b
|
||||
size 190687
|
||||
3
assets/main-results.png
Normal file
3
assets/main-results.png
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:c9add60799cf48907664cb3d672c63b4ba1ed561b1b233998338bb2fe64a8b30
|
||||
size 126061
|
||||
BIN
assets/moe-results.png
Normal file
BIN
assets/moe-results.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 40 KiB |
3
assets/overview.png
Normal file
3
assets/overview.png
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:ccd2cd0cc60cf3ac24bef7e473d11325141374803154fced548d4530a914acf4
|
||||
size 150969
|
||||
BIN
assets/teaser.png
Normal file
BIN
assets/teaser.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 83 KiB |
89
chat_template.jinja
Normal file
89
chat_template.jinja
Normal file
@@ -0,0 +1,89 @@
|
||||
{%- if tools %}
|
||||
{{- '<|im_start|>system\n' }}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- messages[0].content + '\n\n' }}
|
||||
{%- endif %}
|
||||
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
||||
{%- for tool in tools %}
|
||||
{{- "\n" }}
|
||||
{{- tool | tojson }}
|
||||
{%- endfor %}
|
||||
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
||||
{%- else %}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
||||
{%- for message in messages[::-1] %}
|
||||
{%- set index = (messages|length - 1) - loop.index0 %}
|
||||
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
||||
{%- set ns.multi_step_tool = false %}
|
||||
{%- set ns.last_query_index = index %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- for message in messages %}
|
||||
{%- if message.content is string %}
|
||||
{%- set content = message.content %}
|
||||
{%- else %}
|
||||
{%- set content = '' %}
|
||||
{%- endif %}
|
||||
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
||||
{%- elif message.role == "assistant" %}
|
||||
{%- set reasoning_content = '' %}
|
||||
{%- if message.reasoning_content is string %}
|
||||
{%- set reasoning_content = message.reasoning_content %}
|
||||
{%- else %}
|
||||
{%- if '</think>' in content %}
|
||||
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
||||
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- if loop.index0 > ns.last_query_index %}
|
||||
{%- if loop.last or (not loop.last and reasoning_content) %}
|
||||
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- if message.tool_calls %}
|
||||
{%- for tool_call in message.tool_calls %}
|
||||
{%- if (loop.first and content) or (not loop.first) %}
|
||||
{{- '\n' }}
|
||||
{%- endif %}
|
||||
{%- if tool_call.function %}
|
||||
{%- set tool_call = tool_call.function %}
|
||||
{%- endif %}
|
||||
{{- '<tool_call>\n{"name": "' }}
|
||||
{{- tool_call.name }}
|
||||
{{- '", "arguments": ' }}
|
||||
{%- if tool_call.arguments is string %}
|
||||
{{- tool_call.arguments }}
|
||||
{%- else %}
|
||||
{{- tool_call.arguments | tojson }}
|
||||
{%- endif %}
|
||||
{{- '}\n</tool_call>' }}
|
||||
{%- endfor %}
|
||||
{%- endif %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- elif message.role == "tool" %}
|
||||
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
||||
{{- '<|im_start|>user' }}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_response>\n' }}
|
||||
{{- content }}
|
||||
{{- '\n</tool_response>' }}
|
||||
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- if add_generation_prompt %}
|
||||
{{- '<|im_start|>assistant\n' }}
|
||||
{%- if enable_thinking is defined and enable_thinking is false %}
|
||||
{{- '<think>\n\n</think>\n\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
61
checkpoints/qwen3-4b-base-sft-qwen3-8b/README.md
Normal file
61
checkpoints/qwen3-4b-base-sft-qwen3-8b/README.md
Normal file
@@ -0,0 +1,61 @@
|
||||
---
|
||||
library_name: transformers
|
||||
license: other
|
||||
base_model: model_weights/qwen3-4b-base
|
||||
tags:
|
||||
- llama-factory
|
||||
- full
|
||||
- generated_from_trainer
|
||||
model-index:
|
||||
- name: qwen3-4b-base-sft-qwen3-8b
|
||||
results: []
|
||||
---
|
||||
|
||||
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
||||
should probably proofread and complete it, then remove this comment. -->
|
||||
|
||||
# qwen3-4b-base-sft-qwen3-8b
|
||||
|
||||
This model is a fine-tuned version of [model_weights/qwen3-4b-base](https://huggingface.co/model_weights/qwen3-4b-base) on the openthoughts3_300k_qwen3-8b dataset.
|
||||
|
||||
## Model description
|
||||
|
||||
More information needed
|
||||
|
||||
## Intended uses & limitations
|
||||
|
||||
More information needed
|
||||
|
||||
## Training and evaluation data
|
||||
|
||||
More information needed
|
||||
|
||||
## Training procedure
|
||||
|
||||
### Training hyperparameters
|
||||
|
||||
The following hyperparameters were used during training:
|
||||
- learning_rate: 8e-05
|
||||
- train_batch_size: 4
|
||||
- eval_batch_size: 8
|
||||
- seed: 42
|
||||
- distributed_type: multi-GPU
|
||||
- num_devices: 4
|
||||
- gradient_accumulation_steps: 2
|
||||
- total_train_batch_size: 32
|
||||
- total_eval_batch_size: 32
|
||||
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
||||
- lr_scheduler_type: cosine
|
||||
- lr_scheduler_warmup_ratio: 0.1
|
||||
- training_steps: 3000
|
||||
|
||||
### Training results
|
||||
|
||||
|
||||
|
||||
### Framework versions
|
||||
|
||||
- Transformers 4.52.4
|
||||
- Pytorch 2.12.0+cu130
|
||||
- Datasets 3.6.0
|
||||
- Tokenizers 0.21.1
|
||||
28
checkpoints/qwen3-4b-base-sft-qwen3-8b/added_tokens.json
Normal file
28
checkpoints/qwen3-4b-base-sft-qwen3-8b/added_tokens.json
Normal file
@@ -0,0 +1,28 @@
|
||||
{
|
||||
"</think>": 151668,
|
||||
"</tool_call>": 151658,
|
||||
"</tool_response>": 151666,
|
||||
"<think>": 151667,
|
||||
"<tool_call>": 151657,
|
||||
"<tool_response>": 151665,
|
||||
"<|box_end|>": 151649,
|
||||
"<|box_start|>": 151648,
|
||||
"<|endoftext|>": 151643,
|
||||
"<|file_sep|>": 151664,
|
||||
"<|fim_middle|>": 151660,
|
||||
"<|fim_pad|>": 151662,
|
||||
"<|fim_prefix|>": 151659,
|
||||
"<|fim_suffix|>": 151661,
|
||||
"<|im_end|>": 151645,
|
||||
"<|im_start|>": 151644,
|
||||
"<|image_pad|>": 151655,
|
||||
"<|object_ref_end|>": 151647,
|
||||
"<|object_ref_start|>": 151646,
|
||||
"<|quad_end|>": 151651,
|
||||
"<|quad_start|>": 151650,
|
||||
"<|repo_name|>": 151663,
|
||||
"<|video_pad|>": 151656,
|
||||
"<|vision_end|>": 151653,
|
||||
"<|vision_pad|>": 151654,
|
||||
"<|vision_start|>": 151652
|
||||
}
|
||||
8
checkpoints/qwen3-4b-base-sft-qwen3-8b/all_results.json
Normal file
8
checkpoints/qwen3-4b-base-sft-qwen3-8b/all_results.json
Normal file
@@ -0,0 +1,8 @@
|
||||
{
|
||||
"epoch": 0.34443168771526983,
|
||||
"total_flos": 3.429011909561549e+19,
|
||||
"train_loss": 0.27564545996983847,
|
||||
"train_runtime": 45365.0148,
|
||||
"train_samples_per_second": 2.116,
|
||||
"train_steps_per_second": 0.066
|
||||
}
|
||||
89
checkpoints/qwen3-4b-base-sft-qwen3-8b/chat_template.jinja
Normal file
89
checkpoints/qwen3-4b-base-sft-qwen3-8b/chat_template.jinja
Normal file
@@ -0,0 +1,89 @@
|
||||
{%- if tools %}
|
||||
{{- '<|im_start|>system\n' }}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- messages[0].content + '\n\n' }}
|
||||
{%- endif %}
|
||||
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
||||
{%- for tool in tools %}
|
||||
{{- "\n" }}
|
||||
{{- tool | tojson }}
|
||||
{%- endfor %}
|
||||
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
||||
{%- else %}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
||||
{%- for message in messages[::-1] %}
|
||||
{%- set index = (messages|length - 1) - loop.index0 %}
|
||||
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
||||
{%- set ns.multi_step_tool = false %}
|
||||
{%- set ns.last_query_index = index %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- for message in messages %}
|
||||
{%- if message.content is string %}
|
||||
{%- set content = message.content %}
|
||||
{%- else %}
|
||||
{%- set content = '' %}
|
||||
{%- endif %}
|
||||
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
||||
{%- elif message.role == "assistant" %}
|
||||
{%- set reasoning_content = '' %}
|
||||
{%- if message.reasoning_content is string %}
|
||||
{%- set reasoning_content = message.reasoning_content %}
|
||||
{%- else %}
|
||||
{%- if '</think>' in content %}
|
||||
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
||||
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- if loop.index0 > ns.last_query_index %}
|
||||
{%- if loop.last or (not loop.last and reasoning_content) %}
|
||||
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- if message.tool_calls %}
|
||||
{%- for tool_call in message.tool_calls %}
|
||||
{%- if (loop.first and content) or (not loop.first) %}
|
||||
{{- '\n' }}
|
||||
{%- endif %}
|
||||
{%- if tool_call.function %}
|
||||
{%- set tool_call = tool_call.function %}
|
||||
{%- endif %}
|
||||
{{- '<tool_call>\n{"name": "' }}
|
||||
{{- tool_call.name }}
|
||||
{{- '", "arguments": ' }}
|
||||
{%- if tool_call.arguments is string %}
|
||||
{{- tool_call.arguments }}
|
||||
{%- else %}
|
||||
{{- tool_call.arguments | tojson }}
|
||||
{%- endif %}
|
||||
{{- '}\n</tool_call>' }}
|
||||
{%- endfor %}
|
||||
{%- endif %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- elif message.role == "tool" %}
|
||||
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
||||
{{- '<|im_start|>user' }}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_response>\n' }}
|
||||
{{- content }}
|
||||
{{- '\n</tool_response>' }}
|
||||
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- if add_generation_prompt %}
|
||||
{{- '<|im_start|>assistant\n' }}
|
||||
{%- if enable_thinking is defined and enable_thinking is false %}
|
||||
{{- '<think>\n\n</think>\n\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
30
checkpoints/qwen3-4b-base-sft-qwen3-8b/config.json
Normal file
30
checkpoints/qwen3-4b-base-sft-qwen3-8b/config.json
Normal file
@@ -0,0 +1,30 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3ForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": 151643,
|
||||
"eos_token_id": 151645,
|
||||
"head_dim": 128,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 2560,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 9728,
|
||||
"max_position_embeddings": 40960,
|
||||
"max_window_layers": 36,
|
||||
"model_type": "qwen3",
|
||||
"num_attention_heads": 32,
|
||||
"num_hidden_layers": 36,
|
||||
"num_key_value_heads": 8,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_scaling": null,
|
||||
"rope_theta": 1000000,
|
||||
"sliding_window": null,
|
||||
"tie_word_embeddings": true,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.52.4",
|
||||
"use_cache": false,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936
|
||||
}
|
||||
@@ -0,0 +1,13 @@
|
||||
{
|
||||
"bos_token_id": 151643,
|
||||
"do_sample": true,
|
||||
"eos_token_id": [
|
||||
151645,
|
||||
151643
|
||||
],
|
||||
"pad_token_id": 151643,
|
||||
"temperature": 0.6,
|
||||
"top_k": 20,
|
||||
"top_p": 0.95,
|
||||
"transformers_version": "4.52.4"
|
||||
}
|
||||
151388
checkpoints/qwen3-4b-base-sft-qwen3-8b/merges.txt
Normal file
151388
checkpoints/qwen3-4b-base-sft-qwen3-8b/merges.txt
Normal file
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:3d57b393ee9734c673d28da679609d593ade800fb4dfc7774a388da2a0be5654
|
||||
size 4967215360
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:2fd291e26278a84c71130076fd0ce28ed048b625e6725e893fdaf4a9dc6c40ee
|
||||
size 3855679144
|
||||
@@ -0,0 +1,406 @@
|
||||
{
|
||||
"metadata": {
|
||||
"total_size": 8822848512
|
||||
},
|
||||
"weight_map": {
|
||||
"lm_head.weight": "model-00002-of-00002.safetensors",
|
||||
"model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.20.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.norm.weight": "model-00002-of-00002.safetensors"
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,31 @@
|
||||
{
|
||||
"additional_special_tokens": [
|
||||
"<|im_start|>",
|
||||
"<|im_end|>",
|
||||
"<|object_ref_start|>",
|
||||
"<|object_ref_end|>",
|
||||
"<|box_start|>",
|
||||
"<|box_end|>",
|
||||
"<|quad_start|>",
|
||||
"<|quad_end|>",
|
||||
"<|vision_start|>",
|
||||
"<|vision_end|>",
|
||||
"<|vision_pad|>",
|
||||
"<|image_pad|>",
|
||||
"<|video_pad|>"
|
||||
],
|
||||
"eos_token": {
|
||||
"content": "<|im_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
},
|
||||
"pad_token": {
|
||||
"content": "<|endoftext|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
}
|
||||
}
|
||||
BIN
checkpoints/qwen3-4b-base-sft-qwen3-8b/tokenizer.json
(Stored with Git LFS)
Normal file
BIN
checkpoints/qwen3-4b-base-sft-qwen3-8b/tokenizer.json
(Stored with Git LFS)
Normal file
Binary file not shown.
240
checkpoints/qwen3-4b-base-sft-qwen3-8b/tokenizer_config.json
Normal file
240
checkpoints/qwen3-4b-base-sft-qwen3-8b/tokenizer_config.json
Normal file
@@ -0,0 +1,240 @@
|
||||
{
|
||||
"add_bos_token": false,
|
||||
"add_prefix_space": false,
|
||||
"added_tokens_decoder": {
|
||||
"151643": {
|
||||
"content": "<|endoftext|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151644": {
|
||||
"content": "<|im_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151645": {
|
||||
"content": "<|im_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151646": {
|
||||
"content": "<|object_ref_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151647": {
|
||||
"content": "<|object_ref_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151648": {
|
||||
"content": "<|box_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151649": {
|
||||
"content": "<|box_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151650": {
|
||||
"content": "<|quad_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151651": {
|
||||
"content": "<|quad_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151652": {
|
||||
"content": "<|vision_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151653": {
|
||||
"content": "<|vision_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151654": {
|
||||
"content": "<|vision_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151655": {
|
||||
"content": "<|image_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151656": {
|
||||
"content": "<|video_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151657": {
|
||||
"content": "<tool_call>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151658": {
|
||||
"content": "</tool_call>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151659": {
|
||||
"content": "<|fim_prefix|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151660": {
|
||||
"content": "<|fim_middle|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151661": {
|
||||
"content": "<|fim_suffix|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151662": {
|
||||
"content": "<|fim_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151663": {
|
||||
"content": "<|repo_name|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151664": {
|
||||
"content": "<|file_sep|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151665": {
|
||||
"content": "<tool_response>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151666": {
|
||||
"content": "</tool_response>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151667": {
|
||||
"content": "<think>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151668": {
|
||||
"content": "</think>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
}
|
||||
},
|
||||
"additional_special_tokens": [
|
||||
"<|im_start|>",
|
||||
"<|im_end|>",
|
||||
"<|object_ref_start|>",
|
||||
"<|object_ref_end|>",
|
||||
"<|box_start|>",
|
||||
"<|box_end|>",
|
||||
"<|quad_start|>",
|
||||
"<|quad_end|>",
|
||||
"<|vision_start|>",
|
||||
"<|vision_end|>",
|
||||
"<|vision_pad|>",
|
||||
"<|image_pad|>",
|
||||
"<|video_pad|>"
|
||||
],
|
||||
"bos_token": null,
|
||||
"clean_up_tokenization_spaces": false,
|
||||
"eos_token": "<|im_end|>",
|
||||
"errors": "replace",
|
||||
"extra_special_tokens": {},
|
||||
"model_max_length": 131072,
|
||||
"pad_token": "<|endoftext|>",
|
||||
"padding_side": "right",
|
||||
"split_special_tokens": false,
|
||||
"tokenizer_class": "Qwen2Tokenizer",
|
||||
"unk_token": null
|
||||
}
|
||||
@@ -0,0 +1,8 @@
|
||||
{
|
||||
"epoch": 0.34443168771526983,
|
||||
"total_flos": 3.429011909561549e+19,
|
||||
"train_loss": 0.27564545996983847,
|
||||
"train_runtime": 45365.0148,
|
||||
"train_samples_per_second": 2.116,
|
||||
"train_steps_per_second": 0.066
|
||||
}
|
||||
3001
checkpoints/qwen3-4b-base-sft-qwen3-8b/trainer_log.jsonl
Normal file
3001
checkpoints/qwen3-4b-base-sft-qwen3-8b/trainer_log.jsonl
Normal file
File diff suppressed because it is too large
Load Diff
21043
checkpoints/qwen3-4b-base-sft-qwen3-8b/trainer_state.json
Normal file
21043
checkpoints/qwen3-4b-base-sft-qwen3-8b/trainer_state.json
Normal file
File diff suppressed because it is too large
Load Diff
3
checkpoints/qwen3-4b-base-sft-qwen3-8b/training_args.bin
Normal file
3
checkpoints/qwen3-4b-base-sft-qwen3-8b/training_args.bin
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:b28ccff420c215e9d7eaa03851e14301ec0e1720a9cca5e39fab81383ec4e4a7
|
||||
size 7377
|
||||
BIN
checkpoints/qwen3-4b-base-sft-qwen3-8b/training_loss.png
Normal file
BIN
checkpoints/qwen3-4b-base-sft-qwen3-8b/training_loss.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 41 KiB |
1
checkpoints/qwen3-4b-base-sft-qwen3-8b/vocab.json
Normal file
1
checkpoints/qwen3-4b-base-sft-qwen3-8b/vocab.json
Normal file
File diff suppressed because one or more lines are too long
61
checkpoints/qwen3-4b-lightning-opd-hf/README.md
Normal file
61
checkpoints/qwen3-4b-lightning-opd-hf/README.md
Normal file
@@ -0,0 +1,61 @@
|
||||
---
|
||||
library_name: transformers
|
||||
license: other
|
||||
base_model: model_weights/qwen3-4b-base
|
||||
tags:
|
||||
- llama-factory
|
||||
- full
|
||||
- generated_from_trainer
|
||||
model-index:
|
||||
- name: qwen3-4b-base-sft-qwen3-8b
|
||||
results: []
|
||||
---
|
||||
|
||||
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
||||
should probably proofread and complete it, then remove this comment. -->
|
||||
|
||||
# qwen3-4b-base-sft-qwen3-8b
|
||||
|
||||
This model is a fine-tuned version of [model_weights/qwen3-4b-base](https://huggingface.co/model_weights/qwen3-4b-base) on the openthoughts3_300k_qwen3-8b dataset.
|
||||
|
||||
## Model description
|
||||
|
||||
More information needed
|
||||
|
||||
## Intended uses & limitations
|
||||
|
||||
More information needed
|
||||
|
||||
## Training and evaluation data
|
||||
|
||||
More information needed
|
||||
|
||||
## Training procedure
|
||||
|
||||
### Training hyperparameters
|
||||
|
||||
The following hyperparameters were used during training:
|
||||
- learning_rate: 8e-05
|
||||
- train_batch_size: 4
|
||||
- eval_batch_size: 8
|
||||
- seed: 42
|
||||
- distributed_type: multi-GPU
|
||||
- num_devices: 4
|
||||
- gradient_accumulation_steps: 2
|
||||
- total_train_batch_size: 32
|
||||
- total_eval_batch_size: 32
|
||||
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
|
||||
- lr_scheduler_type: cosine
|
||||
- lr_scheduler_warmup_ratio: 0.1
|
||||
- training_steps: 3000
|
||||
|
||||
### Training results
|
||||
|
||||
|
||||
|
||||
### Framework versions
|
||||
|
||||
- Transformers 4.52.4
|
||||
- Pytorch 2.12.0+cu130
|
||||
- Datasets 3.6.0
|
||||
- Tokenizers 0.21.1
|
||||
28
checkpoints/qwen3-4b-lightning-opd-hf/added_tokens.json
Normal file
28
checkpoints/qwen3-4b-lightning-opd-hf/added_tokens.json
Normal file
@@ -0,0 +1,28 @@
|
||||
{
|
||||
"</think>": 151668,
|
||||
"</tool_call>": 151658,
|
||||
"</tool_response>": 151666,
|
||||
"<think>": 151667,
|
||||
"<tool_call>": 151657,
|
||||
"<tool_response>": 151665,
|
||||
"<|box_end|>": 151649,
|
||||
"<|box_start|>": 151648,
|
||||
"<|endoftext|>": 151643,
|
||||
"<|file_sep|>": 151664,
|
||||
"<|fim_middle|>": 151660,
|
||||
"<|fim_pad|>": 151662,
|
||||
"<|fim_prefix|>": 151659,
|
||||
"<|fim_suffix|>": 151661,
|
||||
"<|im_end|>": 151645,
|
||||
"<|im_start|>": 151644,
|
||||
"<|image_pad|>": 151655,
|
||||
"<|object_ref_end|>": 151647,
|
||||
"<|object_ref_start|>": 151646,
|
||||
"<|quad_end|>": 151651,
|
||||
"<|quad_start|>": 151650,
|
||||
"<|repo_name|>": 151663,
|
||||
"<|video_pad|>": 151656,
|
||||
"<|vision_end|>": 151653,
|
||||
"<|vision_pad|>": 151654,
|
||||
"<|vision_start|>": 151652
|
||||
}
|
||||
8
checkpoints/qwen3-4b-lightning-opd-hf/all_results.json
Normal file
8
checkpoints/qwen3-4b-lightning-opd-hf/all_results.json
Normal file
@@ -0,0 +1,8 @@
|
||||
{
|
||||
"epoch": 0.34443168771526983,
|
||||
"total_flos": 3.429011909561549e+19,
|
||||
"train_loss": 0.27564545996983847,
|
||||
"train_runtime": 45365.0148,
|
||||
"train_samples_per_second": 2.116,
|
||||
"train_steps_per_second": 0.066
|
||||
}
|
||||
89
checkpoints/qwen3-4b-lightning-opd-hf/chat_template.jinja
Normal file
89
checkpoints/qwen3-4b-lightning-opd-hf/chat_template.jinja
Normal file
@@ -0,0 +1,89 @@
|
||||
{%- if tools %}
|
||||
{{- '<|im_start|>system\n' }}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- messages[0].content + '\n\n' }}
|
||||
{%- endif %}
|
||||
{{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
|
||||
{%- for tool in tools %}
|
||||
{{- "\n" }}
|
||||
{{- tool | tojson }}
|
||||
{%- endfor %}
|
||||
{{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
|
||||
{%- else %}
|
||||
{%- if messages[0].role == 'system' %}
|
||||
{{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
|
||||
{%- for message in messages[::-1] %}
|
||||
{%- set index = (messages|length - 1) - loop.index0 %}
|
||||
{%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
|
||||
{%- set ns.multi_step_tool = false %}
|
||||
{%- set ns.last_query_index = index %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- for message in messages %}
|
||||
{%- if message.content is string %}
|
||||
{%- set content = message.content %}
|
||||
{%- else %}
|
||||
{%- set content = '' %}
|
||||
{%- endif %}
|
||||
{%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
|
||||
{%- elif message.role == "assistant" %}
|
||||
{%- set reasoning_content = '' %}
|
||||
{%- if message.reasoning_content is string %}
|
||||
{%- set reasoning_content = message.reasoning_content %}
|
||||
{%- else %}
|
||||
{%- if '</think>' in content %}
|
||||
{%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
|
||||
{%- set content = content.split('</think>')[-1].lstrip('\n') %}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- if loop.index0 > ns.last_query_index %}
|
||||
{%- if loop.last or (not loop.last and reasoning_content) %}
|
||||
{{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + content }}
|
||||
{%- endif %}
|
||||
{%- if message.tool_calls %}
|
||||
{%- for tool_call in message.tool_calls %}
|
||||
{%- if (loop.first and content) or (not loop.first) %}
|
||||
{{- '\n' }}
|
||||
{%- endif %}
|
||||
{%- if tool_call.function %}
|
||||
{%- set tool_call = tool_call.function %}
|
||||
{%- endif %}
|
||||
{{- '<tool_call>\n{"name": "' }}
|
||||
{{- tool_call.name }}
|
||||
{{- '", "arguments": ' }}
|
||||
{%- if tool_call.arguments is string %}
|
||||
{{- tool_call.arguments }}
|
||||
{%- else %}
|
||||
{{- tool_call.arguments | tojson }}
|
||||
{%- endif %}
|
||||
{{- '}\n</tool_call>' }}
|
||||
{%- endfor %}
|
||||
{%- endif %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- elif message.role == "tool" %}
|
||||
{%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
|
||||
{{- '<|im_start|>user' }}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_response>\n' }}
|
||||
{{- content }}
|
||||
{{- '\n</tool_response>' }}
|
||||
{%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- endfor %}
|
||||
{%- if add_generation_prompt %}
|
||||
{{- '<|im_start|>assistant\n' }}
|
||||
{%- if enable_thinking is defined and enable_thinking is false %}
|
||||
{{- '<think>\n\n</think>\n\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
30
checkpoints/qwen3-4b-lightning-opd-hf/config.json
Normal file
30
checkpoints/qwen3-4b-lightning-opd-hf/config.json
Normal file
@@ -0,0 +1,30 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3ForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": 151643,
|
||||
"eos_token_id": 151645,
|
||||
"head_dim": 128,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 2560,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 9728,
|
||||
"max_position_embeddings": 40960,
|
||||
"max_window_layers": 36,
|
||||
"model_type": "qwen3",
|
||||
"num_attention_heads": 32,
|
||||
"num_hidden_layers": 36,
|
||||
"num_key_value_heads": 8,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_scaling": null,
|
||||
"rope_theta": 1000000,
|
||||
"sliding_window": null,
|
||||
"tie_word_embeddings": true,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.52.4",
|
||||
"use_cache": false,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936
|
||||
}
|
||||
13
checkpoints/qwen3-4b-lightning-opd-hf/generation_config.json
Normal file
13
checkpoints/qwen3-4b-lightning-opd-hf/generation_config.json
Normal file
@@ -0,0 +1,13 @@
|
||||
{
|
||||
"bos_token_id": 151643,
|
||||
"do_sample": true,
|
||||
"eos_token_id": [
|
||||
151645,
|
||||
151643
|
||||
],
|
||||
"pad_token_id": 151643,
|
||||
"temperature": 0.6,
|
||||
"top_k": 20,
|
||||
"top_p": 0.95,
|
||||
"transformers_version": "4.52.4"
|
||||
}
|
||||
151388
checkpoints/qwen3-4b-lightning-opd-hf/merges.txt
Normal file
151388
checkpoints/qwen3-4b-lightning-opd-hf/merges.txt
Normal file
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:9347d0d8aeae95b73ee3af7175c658679b6e4544a6f1c65a37c945520ee86a0e
|
||||
size 5355373240
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:a926c6e2698cff38084f84796a6fe6e41e3bf83961e34f960eac18cb00b4bd74
|
||||
size 2689608768
|
||||
@@ -0,0 +1,405 @@
|
||||
{
|
||||
"metadata": {
|
||||
"total_size": 8044936192
|
||||
},
|
||||
"weight_map": {
|
||||
"model.embed_tokens.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.o_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.input_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.q_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.k_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.v_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.q_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.k_norm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.post_attention_layernorm.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.10.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.11.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.12.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.13.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.14.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.15.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.16.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.17.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.18.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.19.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.20.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.21.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.22.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.23.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.24.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.25.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.26.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.27.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.28.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.29.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.30.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.31.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.32.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.33.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.34.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.35.mlp.down_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.0.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.1.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.2.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.3.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.4.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.5.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.6.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.7.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.mlp.gate_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.8.mlp.up_proj.weight": "model-00000-of-00002.safetensors",
|
||||
"model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.21.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.21.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.22.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.22.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.23.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.23.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.24.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.24.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.25.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.25.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.26.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.26.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.27.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.27.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.28.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.28.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.29.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.29.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.30.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.30.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.31.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.31.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.32.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.32.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.33.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.33.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.34.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.34.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.35.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.35.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.norm.weight": "model-00001-of-00002.safetensors"
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,31 @@
|
||||
{
|
||||
"additional_special_tokens": [
|
||||
"<|im_start|>",
|
||||
"<|im_end|>",
|
||||
"<|object_ref_start|>",
|
||||
"<|object_ref_end|>",
|
||||
"<|box_start|>",
|
||||
"<|box_end|>",
|
||||
"<|quad_start|>",
|
||||
"<|quad_end|>",
|
||||
"<|vision_start|>",
|
||||
"<|vision_end|>",
|
||||
"<|vision_pad|>",
|
||||
"<|image_pad|>",
|
||||
"<|video_pad|>"
|
||||
],
|
||||
"eos_token": {
|
||||
"content": "<|im_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
},
|
||||
"pad_token": {
|
||||
"content": "<|endoftext|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false
|
||||
}
|
||||
}
|
||||
BIN
checkpoints/qwen3-4b-lightning-opd-hf/tokenizer.json
(Stored with Git LFS)
Normal file
BIN
checkpoints/qwen3-4b-lightning-opd-hf/tokenizer.json
(Stored with Git LFS)
Normal file
Binary file not shown.
240
checkpoints/qwen3-4b-lightning-opd-hf/tokenizer_config.json
Normal file
240
checkpoints/qwen3-4b-lightning-opd-hf/tokenizer_config.json
Normal file
@@ -0,0 +1,240 @@
|
||||
{
|
||||
"add_bos_token": false,
|
||||
"add_prefix_space": false,
|
||||
"added_tokens_decoder": {
|
||||
"151643": {
|
||||
"content": "<|endoftext|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151644": {
|
||||
"content": "<|im_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151645": {
|
||||
"content": "<|im_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151646": {
|
||||
"content": "<|object_ref_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151647": {
|
||||
"content": "<|object_ref_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151648": {
|
||||
"content": "<|box_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151649": {
|
||||
"content": "<|box_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151650": {
|
||||
"content": "<|quad_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151651": {
|
||||
"content": "<|quad_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151652": {
|
||||
"content": "<|vision_start|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151653": {
|
||||
"content": "<|vision_end|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151654": {
|
||||
"content": "<|vision_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151655": {
|
||||
"content": "<|image_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151656": {
|
||||
"content": "<|video_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": true
|
||||
},
|
||||
"151657": {
|
||||
"content": "<tool_call>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151658": {
|
||||
"content": "</tool_call>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151659": {
|
||||
"content": "<|fim_prefix|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151660": {
|
||||
"content": "<|fim_middle|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151661": {
|
||||
"content": "<|fim_suffix|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151662": {
|
||||
"content": "<|fim_pad|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151663": {
|
||||
"content": "<|repo_name|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151664": {
|
||||
"content": "<|file_sep|>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151665": {
|
||||
"content": "<tool_response>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151666": {
|
||||
"content": "</tool_response>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151667": {
|
||||
"content": "<think>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
},
|
||||
"151668": {
|
||||
"content": "</think>",
|
||||
"lstrip": false,
|
||||
"normalized": false,
|
||||
"rstrip": false,
|
||||
"single_word": false,
|
||||
"special": false
|
||||
}
|
||||
},
|
||||
"additional_special_tokens": [
|
||||
"<|im_start|>",
|
||||
"<|im_end|>",
|
||||
"<|object_ref_start|>",
|
||||
"<|object_ref_end|>",
|
||||
"<|box_start|>",
|
||||
"<|box_end|>",
|
||||
"<|quad_start|>",
|
||||
"<|quad_end|>",
|
||||
"<|vision_start|>",
|
||||
"<|vision_end|>",
|
||||
"<|vision_pad|>",
|
||||
"<|image_pad|>",
|
||||
"<|video_pad|>"
|
||||
],
|
||||
"bos_token": null,
|
||||
"clean_up_tokenization_spaces": false,
|
||||
"eos_token": "<|im_end|>",
|
||||
"errors": "replace",
|
||||
"extra_special_tokens": {},
|
||||
"model_max_length": 131072,
|
||||
"pad_token": "<|endoftext|>",
|
||||
"padding_side": "right",
|
||||
"split_special_tokens": false,
|
||||
"tokenizer_class": "Qwen2Tokenizer",
|
||||
"unk_token": null
|
||||
}
|
||||
8
checkpoints/qwen3-4b-lightning-opd-hf/train_results.json
Normal file
8
checkpoints/qwen3-4b-lightning-opd-hf/train_results.json
Normal file
@@ -0,0 +1,8 @@
|
||||
{
|
||||
"epoch": 0.34443168771526983,
|
||||
"total_flos": 3.429011909561549e+19,
|
||||
"train_loss": 0.27564545996983847,
|
||||
"train_runtime": 45365.0148,
|
||||
"train_samples_per_second": 2.116,
|
||||
"train_steps_per_second": 0.066
|
||||
}
|
||||
3001
checkpoints/qwen3-4b-lightning-opd-hf/trainer_log.jsonl
Normal file
3001
checkpoints/qwen3-4b-lightning-opd-hf/trainer_log.jsonl
Normal file
File diff suppressed because it is too large
Load Diff
21043
checkpoints/qwen3-4b-lightning-opd-hf/trainer_state.json
Normal file
21043
checkpoints/qwen3-4b-lightning-opd-hf/trainer_state.json
Normal file
File diff suppressed because it is too large
Load Diff
3
checkpoints/qwen3-4b-lightning-opd-hf/training_args.bin
Normal file
3
checkpoints/qwen3-4b-lightning-opd-hf/training_args.bin
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:b28ccff420c215e9d7eaa03851e14301ec0e1720a9cca5e39fab81383ec4e4a7
|
||||
size 7377
|
||||
BIN
checkpoints/qwen3-4b-lightning-opd-hf/training_loss.png
Normal file
BIN
checkpoints/qwen3-4b-lightning-opd-hf/training_loss.png
Normal file
Binary file not shown.
|
After Width: | Height: | Size: 41 KiB |
1
checkpoints/qwen3-4b-lightning-opd-hf/vocab.json
Normal file
1
checkpoints/qwen3-4b-lightning-opd-hf/vocab.json
Normal file
File diff suppressed because one or more lines are too long
30
config.json
Normal file
30
config.json
Normal file
@@ -0,0 +1,30 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen3ForCausalLM"
|
||||
],
|
||||
"attention_bias": false,
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": 151643,
|
||||
"eos_token_id": 151645,
|
||||
"head_dim": 128,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 2560,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 9728,
|
||||
"max_position_embeddings": 40960,
|
||||
"max_window_layers": 36,
|
||||
"model_type": "qwen3",
|
||||
"num_attention_heads": 32,
|
||||
"num_hidden_layers": 36,
|
||||
"num_key_value_heads": 8,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_scaling": null,
|
||||
"rope_theta": 1000000,
|
||||
"sliding_window": null,
|
||||
"tie_word_embeddings": true,
|
||||
"torch_dtype": "bfloat16",
|
||||
"transformers_version": "4.52.4",
|
||||
"use_cache": false,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936
|
||||
}
|
||||
145
configs/lightning_opd/qwen3-30b-a3b-lightning-opd.py
Normal file
145
configs/lightning_opd/qwen3-30b-a3b-lightning-opd.py
Normal file
@@ -0,0 +1,145 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import slime.utils.external_utils.command_utils as U
|
||||
|
||||
# Lightning OPD: response tokens + teacher logprobs are pre-computed in parquet.
|
||||
# No teacher server needed at training time -> all 8 GPUs go to the actor.
|
||||
#
|
||||
# Required env vars:
|
||||
# SFT_CHECKPOINT - path to the SFT checkpoint (HF format)
|
||||
# LIGHTNING_OPD_DATA - path to the precomputed parquet
|
||||
|
||||
MODEL_NAME = "Qwen3-30B-A3B-Base-Open-Thoughts-Qwen3-30B-A3B-Thinking-2507-sft"
|
||||
MODEL_TYPE = "qwen3-30B-A3B"
|
||||
NUM_GPUS = 8
|
||||
SFT_CHECKPOINT = os.environ["SFT_CHECKPOINT"]
|
||||
|
||||
|
||||
def prepare():
|
||||
U.convert_checkpoint(
|
||||
model_name=MODEL_NAME,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
hf_checkpoint=SFT_CHECKPOINT,
|
||||
)
|
||||
|
||||
|
||||
def execute(rerun=True):
|
||||
load_save_path = f"/root/models/{MODEL_NAME}_ckpt__{Path(__file__).stem}/"
|
||||
|
||||
ckpt_args = (
|
||||
f"--hf-checkpoint {SFT_CHECKPOINT} "
|
||||
f"--ref-load /root/models/{MODEL_NAME}_torch_dist "
|
||||
f"--load {load_save_path} "
|
||||
f"--save {load_save_path} "
|
||||
"--save-interval 10 "
|
||||
"--save-retain-interval 100 "
|
||||
)
|
||||
|
||||
rollout_args = (
|
||||
f"--prompt-data {os.environ['LIGHTNING_OPD_DATA']} "
|
||||
"--input-key prompt "
|
||||
"--label-key label "
|
||||
"--rollout-shuffle "
|
||||
"--num-rollout 3000 "
|
||||
"--rollout-batch-size 256 "
|
||||
"--n-samples-per-prompt 1 "
|
||||
"--rollout-max-response-len 4096 "
|
||||
"--global-batch-size 256 "
|
||||
)
|
||||
|
||||
rm_args = (
|
||||
"--custom-rm-path slime.rollout.on_policy_distillation.reward_func "
|
||||
"--custom-reward-post-process-path slime.rollout.on_policy_distillation.post_process_rewards "
|
||||
"--include-verifiable-reward "
|
||||
)
|
||||
|
||||
perf_args = (
|
||||
"--tensor-model-parallel-size 4 "
|
||||
"--sequence-parallel "
|
||||
"--pipeline-model-parallel-size 1 "
|
||||
"--context-parallel-size 1 "
|
||||
"--expert-model-parallel-size 8 "
|
||||
"--expert-tensor-parallel-size 1 "
|
||||
"--recompute-granularity full "
|
||||
"--recompute-method uniform "
|
||||
"--recompute-num-layers 1 "
|
||||
"--use-dynamic-batch-size "
|
||||
"--max-tokens-per-gpu 12288 "
|
||||
)
|
||||
|
||||
grpo_args = (
|
||||
"--advantage-estimator on_policy_distillation "
|
||||
"--use-kl-loss "
|
||||
"--kl-loss-coef 0.00 "
|
||||
"--kl-loss-type low_var_kl "
|
||||
"--entropy-coef 0.00 "
|
||||
)
|
||||
|
||||
optimizer_args = (
|
||||
"--optimizer adam "
|
||||
"--lr 2e-6 "
|
||||
"--lr-decay-style constant "
|
||||
"--weight-decay 0.1 "
|
||||
"--adam-beta1 0.9 "
|
||||
"--adam-beta2 0.98 "
|
||||
"--optimizer-cpu-offload "
|
||||
"--overlap-cpu-optimizer-d2h-h2d "
|
||||
"--use-precision-aware-optimizer "
|
||||
)
|
||||
|
||||
wandb_args = ""
|
||||
if os.environ.get("WANDB_KEY"):
|
||||
wandb_args = (
|
||||
"--use-wandb "
|
||||
"--wandb-project lightning-opd "
|
||||
f"--wandb-group {Path(__file__).stem} "
|
||||
f"--wandb-key {os.environ['WANDB_KEY']} "
|
||||
)
|
||||
|
||||
sglang_args = (
|
||||
"--rollout-num-gpus-per-engine 8 "
|
||||
"--sglang-mem-fraction-static 0.8 "
|
||||
"--sglang-cuda-graph-max-bs 32 "
|
||||
"--sglang-max-running-requests 512 "
|
||||
"--sglang-enable-metrics "
|
||||
)
|
||||
|
||||
misc_args = (
|
||||
"--attention-dropout 0.0 "
|
||||
"--hidden-dropout 0.0 "
|
||||
"--accumulate-allreduce-grads-in-fp32 "
|
||||
"--attention-softmax-in-fp32 "
|
||||
"--attention-backend flash "
|
||||
"--actor-num-nodes 1 "
|
||||
"--actor-num-gpus-per-node 8 "
|
||||
"--rollout-num-gpus 0 "
|
||||
)
|
||||
|
||||
train_args = (
|
||||
f"{ckpt_args} "
|
||||
f"{rollout_args} "
|
||||
f"{rm_args} "
|
||||
f"{grpo_args} "
|
||||
f"{optimizer_args} "
|
||||
f"{wandb_args} "
|
||||
f"{perf_args} "
|
||||
f"{sglang_args} "
|
||||
f"{misc_args} "
|
||||
)
|
||||
|
||||
U.execute_train(
|
||||
rerun=rerun,
|
||||
train_args=train_args,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
prepare()
|
||||
execute(rerun=False)
|
||||
140
configs/lightning_opd/qwen3-4b-lightning-opd.py
Normal file
140
configs/lightning_opd/qwen3-4b-lightning-opd.py
Normal file
@@ -0,0 +1,140 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import slime.utils.external_utils.command_utils as U
|
||||
|
||||
# Lightning OPD: response tokens + teacher logprobs are pre-computed in parquet.
|
||||
# No teacher server needed at training time -> all 8 GPUs go to the actor.
|
||||
#
|
||||
# Required env vars:
|
||||
# SFT_CHECKPOINT - path to the SFT checkpoint (HF format)
|
||||
# LIGHTNING_OPD_DATA - path to the precomputed parquet
|
||||
|
||||
MODEL_NAME = "Qwen3-4B-Base-Open-Thoughts-Qwen3-8B-sft-3k"
|
||||
MODEL_TYPE = "qwen3-4B"
|
||||
NUM_GPUS = 4
|
||||
SFT_CHECKPOINT = os.environ["SFT_CHECKPOINT"]
|
||||
|
||||
|
||||
def prepare():
|
||||
U.convert_checkpoint(
|
||||
model_name=MODEL_NAME,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
hf_checkpoint=SFT_CHECKPOINT,
|
||||
)
|
||||
|
||||
|
||||
def execute(rerun=True):
|
||||
load_save_path = f"/root/models/{MODEL_NAME}_ckpt__{Path(__file__).stem}/"
|
||||
|
||||
ckpt_args = (
|
||||
f"--hf-checkpoint {SFT_CHECKPOINT} "
|
||||
f"--ref-load /root/models/{MODEL_NAME}_torch_dist "
|
||||
f"--load {load_save_path} "
|
||||
f"--save {load_save_path} "
|
||||
"--save-interval 10 "
|
||||
"--save-retain-interval 100 "
|
||||
)
|
||||
|
||||
rollout_args = (
|
||||
f"--prompt-data {os.environ['LIGHTNING_OPD_DATA']} "
|
||||
"--input-key prompt "
|
||||
"--label-key label "
|
||||
"--rollout-shuffle "
|
||||
"--num-rollout 150 "
|
||||
"--rollout-batch-size 256 "
|
||||
"--n-samples-per-prompt 1 "
|
||||
"--rollout-max-response-len 4096 "
|
||||
"--global-batch-size 256 "
|
||||
"--rollout-temperature 0.8"
|
||||
)
|
||||
|
||||
rm_args = (
|
||||
"--custom-rm-path slime.rollout.on_policy_distillation.reward_func "
|
||||
"--custom-reward-post-process-path slime.rollout.on_policy_distillation.post_process_rewards "
|
||||
"--include-verifiable-reward "
|
||||
)
|
||||
|
||||
perf_args = (
|
||||
"--tensor-model-parallel-size 2 "
|
||||
"--sequence-parallel "
|
||||
"--pipeline-model-parallel-size 1 "
|
||||
"--context-parallel-size 1 "
|
||||
"--expert-model-parallel-size 1 "
|
||||
"--expert-tensor-parallel-size 1 "
|
||||
"--recompute-granularity full "
|
||||
"--recompute-method uniform "
|
||||
"--recompute-num-layers 1 "
|
||||
"--use-dynamic-batch-size "
|
||||
"--max-tokens-per-gpu 16384 "
|
||||
)
|
||||
|
||||
grpo_args = (
|
||||
"--advantage-estimator on_policy_distillation "
|
||||
"--use-kl-loss "
|
||||
"--kl-loss-coef 0.00 "
|
||||
"--kl-loss-type low_var_kl "
|
||||
"--entropy-coef 0.00 "
|
||||
)
|
||||
|
||||
optimizer_args = (
|
||||
"--optimizer adam "
|
||||
"--lr 2e-6 "
|
||||
"--lr-decay-style constant "
|
||||
"--weight-decay 0.1 "
|
||||
"--adam-beta1 0.9 "
|
||||
"--adam-beta2 0.98 "
|
||||
)
|
||||
|
||||
wandb_args = ""
|
||||
if os.environ.get("WANDB_KEY"):
|
||||
wandb_args = (
|
||||
"--use-wandb "
|
||||
"--wandb-project lightning-opd "
|
||||
f"--wandb-group {Path(__file__).stem} "
|
||||
f"--wandb-key {os.environ['WANDB_KEY']} "
|
||||
)
|
||||
|
||||
sglang_args = (
|
||||
"--rollout-num-gpus-per-engine 1 "
|
||||
"--sglang-mem-fraction-static 0.4 "
|
||||
)
|
||||
|
||||
misc_args = (
|
||||
"--attention-dropout 0.0 "
|
||||
"--hidden-dropout 0.0 "
|
||||
"--accumulate-allreduce-grads-in-fp32 "
|
||||
"--attention-softmax-in-fp32 "
|
||||
"--attention-backend flash "
|
||||
"--actor-num-nodes 1 "
|
||||
"--actor-num-gpus-per-node 4 "
|
||||
"--rollout-num-gpus 0 "
|
||||
)
|
||||
|
||||
train_args = (
|
||||
f"{ckpt_args} "
|
||||
f"{rollout_args} "
|
||||
f"{rm_args} "
|
||||
f"{grpo_args} "
|
||||
f"{optimizer_args} "
|
||||
f"{wandb_args} "
|
||||
f"{perf_args} "
|
||||
f"{sglang_args} "
|
||||
f"{misc_args} "
|
||||
)
|
||||
|
||||
U.execute_train(
|
||||
rerun=rerun,
|
||||
train_args=train_args,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
prepare()
|
||||
execute(rerun=False)
|
||||
139
configs/lightning_opd/qwen3-8b-lightning-opd.py
Normal file
139
configs/lightning_opd/qwen3-8b-lightning-opd.py
Normal file
@@ -0,0 +1,139 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import slime.utils.external_utils.command_utils as U
|
||||
|
||||
# Lightning OPD: response tokens + teacher logprobs are pre-computed in parquet.
|
||||
# No teacher server needed at training time -> all 8 GPUs go to the actor.
|
||||
#
|
||||
# Required env vars:
|
||||
# SFT_CHECKPOINT - path to the SFT checkpoint (HF format)
|
||||
# LIGHTNING_OPD_DATA - path to the precomputed parquet
|
||||
|
||||
MODEL_NAME = "Qwen3-8B-Base-Open-Thoughts-Qwen3-32B-sft-3k"
|
||||
MODEL_TYPE = "qwen3-8B"
|
||||
NUM_GPUS = 8
|
||||
SFT_CHECKPOINT = os.environ["SFT_CHECKPOINT"]
|
||||
|
||||
|
||||
def prepare():
|
||||
U.convert_checkpoint(
|
||||
model_name=MODEL_NAME,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
hf_checkpoint=SFT_CHECKPOINT,
|
||||
)
|
||||
|
||||
|
||||
def execute(rerun=True):
|
||||
load_save_path = f"/root/models/{MODEL_NAME}_ckpt__{Path(__file__).stem}/"
|
||||
|
||||
ckpt_args = (
|
||||
f"--hf-checkpoint {SFT_CHECKPOINT} "
|
||||
f"--ref-load /root/models/{MODEL_NAME}_torch_dist "
|
||||
f"--load {load_save_path} "
|
||||
f"--save {load_save_path} "
|
||||
"--save-interval 10 "
|
||||
"--save-retain-interval 100 "
|
||||
)
|
||||
|
||||
rollout_args = (
|
||||
f"--prompt-data {os.environ['LIGHTNING_OPD_DATA']} "
|
||||
"--input-key prompt "
|
||||
"--label-key label "
|
||||
"--rollout-shuffle "
|
||||
"--num-rollout 3000 "
|
||||
"--rollout-batch-size 256 "
|
||||
"--n-samples-per-prompt 1 "
|
||||
"--rollout-max-response-len 4096 "
|
||||
"--global-batch-size 256 "
|
||||
)
|
||||
|
||||
rm_args = (
|
||||
"--custom-rm-path slime.rollout.on_policy_distillation.reward_func "
|
||||
"--custom-reward-post-process-path slime.rollout.on_policy_distillation.post_process_rewards "
|
||||
"--include-verifiable-reward "
|
||||
)
|
||||
|
||||
perf_args = (
|
||||
"--tensor-model-parallel-size 4 "
|
||||
"--sequence-parallel "
|
||||
"--pipeline-model-parallel-size 1 "
|
||||
"--context-parallel-size 1 "
|
||||
"--expert-model-parallel-size 1 "
|
||||
"--expert-tensor-parallel-size 1 "
|
||||
"--recompute-granularity full "
|
||||
"--recompute-method uniform "
|
||||
"--recompute-num-layers 1 "
|
||||
"--use-dynamic-batch-size "
|
||||
"--max-tokens-per-gpu 16384 "
|
||||
)
|
||||
|
||||
grpo_args = (
|
||||
"--advantage-estimator on_policy_distillation "
|
||||
"--use-kl-loss "
|
||||
"--kl-loss-coef 0.00 "
|
||||
"--kl-loss-type low_var_kl "
|
||||
"--entropy-coef 0.00 "
|
||||
)
|
||||
|
||||
optimizer_args = (
|
||||
"--optimizer adam "
|
||||
"--lr 2e-6 "
|
||||
"--lr-decay-style constant "
|
||||
"--weight-decay 0.1 "
|
||||
"--adam-beta1 0.9 "
|
||||
"--adam-beta2 0.98 "
|
||||
)
|
||||
|
||||
wandb_args = ""
|
||||
if os.environ.get("WANDB_KEY"):
|
||||
wandb_args = (
|
||||
"--use-wandb "
|
||||
"--wandb-project lightning-opd "
|
||||
f"--wandb-group {Path(__file__).stem} "
|
||||
f"--wandb-key {os.environ['WANDB_KEY']} "
|
||||
)
|
||||
|
||||
sglang_args = (
|
||||
"--rollout-num-gpus-per-engine 1 "
|
||||
"--sglang-mem-fraction-static 0.4 "
|
||||
)
|
||||
|
||||
misc_args = (
|
||||
"--attention-dropout 0.0 "
|
||||
"--hidden-dropout 0.0 "
|
||||
"--accumulate-allreduce-grads-in-fp32 "
|
||||
"--attention-softmax-in-fp32 "
|
||||
"--attention-backend flash "
|
||||
"--actor-num-nodes 1 "
|
||||
"--actor-num-gpus-per-node 8 "
|
||||
"--rollout-num-gpus 0 "
|
||||
)
|
||||
|
||||
train_args = (
|
||||
f"{ckpt_args} "
|
||||
f"{rollout_args} "
|
||||
f"{rm_args} "
|
||||
f"{grpo_args} "
|
||||
f"{optimizer_args} "
|
||||
f"{wandb_args} "
|
||||
f"{perf_args} "
|
||||
f"{sglang_args} "
|
||||
f"{misc_args} "
|
||||
)
|
||||
|
||||
U.execute_train(
|
||||
rerun=rerun,
|
||||
train_args=train_args,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
prepare()
|
||||
execute(rerun=False)
|
||||
49
configs/models/qwen3-30B-A3B.sh
Normal file
49
configs/models/qwen3-30B-A3B.sh
Normal file
@@ -0,0 +1,49 @@
|
||||
NLAYERS=48
|
||||
FIRST_K_DENSE_REPLACE=0
|
||||
|
||||
arr=()
|
||||
for ((i=0; i<NLAYERS; i++)); do
|
||||
if (( i < FIRST_K_DENSE_REPLACE )); then
|
||||
arr+=(0)
|
||||
else
|
||||
arr+=(1)
|
||||
fi
|
||||
done
|
||||
|
||||
printf -v MOE_LAYER_FREQ "[%s]" "$(IFS=', '; echo "${arr[*]}")"
|
||||
|
||||
|
||||
MODEL_ARGS=(
|
||||
--disable-bias-linear
|
||||
--qk-layernorm
|
||||
--group-query-attention
|
||||
--num-attention-heads 32
|
||||
--num-query-groups 4
|
||||
--kv-channels 128
|
||||
--num-layers 48
|
||||
--hidden-size 2048
|
||||
--ffn-hidden-size 6144
|
||||
|
||||
--normalization RMSNorm
|
||||
--position-embedding-type rope
|
||||
--norm-epsilon 1e-6
|
||||
--rotary-percent 1.0
|
||||
--swiglu
|
||||
--untie-embeddings-and-output-weights
|
||||
--vocab-size 151936
|
||||
|
||||
--rotary-base 1000000
|
||||
|
||||
# moe
|
||||
--moe-ffn-hidden-size 768
|
||||
--moe-router-score-function softmax
|
||||
--moe-token-dispatcher-type alltoall
|
||||
--moe-router-topk 8
|
||||
--moe-layer-freq $MOE_LAYER_FREQ
|
||||
--num-experts 128
|
||||
--moe-grouped-gemm
|
||||
--moe-token-drop-policy probs
|
||||
--moe-router-dtype fp32
|
||||
--moe-permute-fusion
|
||||
--moe-aux-loss-coeff 0
|
||||
)
|
||||
20
configs/models/qwen3-4B.sh
Normal file
20
configs/models/qwen3-4B.sh
Normal file
@@ -0,0 +1,20 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
MODEL_ARGS=(
|
||||
--swiglu
|
||||
--num-layers 36
|
||||
--hidden-size 2560
|
||||
--ffn-hidden-size 9728
|
||||
--num-attention-heads 32
|
||||
--group-query-attention
|
||||
--num-query-groups 8
|
||||
--use-rotary-position-embeddings
|
||||
--disable-bias-linear
|
||||
--normalization "RMSNorm"
|
||||
--norm-epsilon 1e-6
|
||||
--rotary-base "${MODEL_ARGS_ROTARY_BASE:-1000000}"
|
||||
--vocab-size 151936
|
||||
--kv-channels 128
|
||||
--qk-layernorm
|
||||
)
|
||||
21
configs/models/qwen3-8B.sh
Normal file
21
configs/models/qwen3-8B.sh
Normal file
@@ -0,0 +1,21 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
MODEL_ARGS=(
|
||||
--swiglu
|
||||
--num-layers 36
|
||||
--hidden-size 4096
|
||||
--ffn-hidden-size 12288
|
||||
--num-attention-heads 32
|
||||
--group-query-attention
|
||||
--num-query-groups 8
|
||||
--use-rotary-position-embeddings
|
||||
--disable-bias-linear
|
||||
--normalization "RMSNorm"
|
||||
--norm-epsilon 1e-6
|
||||
--rotary-base 1000000
|
||||
--vocab-size 151936
|
||||
--kv-channels 128
|
||||
--qk-layernorm
|
||||
--untie-embeddings-and-output-weights
|
||||
)
|
||||
190
configs/opd/qwen3-4b-opd.py
Normal file
190
configs/opd/qwen3-4b-opd.py
Normal file
@@ -0,0 +1,190 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import slime.utils.external_utils.command_utils as U
|
||||
from slime.utils.external_utils.command_utils import get_bool_env_var
|
||||
|
||||
# Standard OPD: requires a live teacher server during training.
|
||||
# 2 GPUs for actor (TP=2), 4 GPUs for rollout, 2 GPUs for teacher server.
|
||||
#
|
||||
# Required env vars:
|
||||
# SFT_CHECKPOINT - path to the SFT checkpoint (HF format)
|
||||
|
||||
MODEL_NAME = "Qwen3-4B-Base-Open-Thoughts-Qwen3-8B-sft-3k"
|
||||
TEACHER_MODEL_NAME = "Qwen3-8B"
|
||||
TEACHER_IP = os.environ.get("MASTER_ADDR", "127.0.0.1")
|
||||
TEACHER_PORT = 13141
|
||||
MODEL_TYPE = "qwen3-4B"
|
||||
NUM_GPUS = 8
|
||||
SFT_CHECKPOINT = os.environ["SFT_CHECKPOINT"]
|
||||
|
||||
|
||||
def deploy_teacher_model():
|
||||
import random, string
|
||||
random_suffix = ''.join(random.choices(string.ascii_letters + string.digits, k=6))
|
||||
LOG_FILE = f"/tmp/sglang_{random_suffix}.log"
|
||||
external_ray = get_bool_env_var("SLIME_SCRIPT_EXTERNAL_RAY")
|
||||
|
||||
U.exec_command(
|
||||
"pkill -9 sglang; "
|
||||
"sleep 3; "
|
||||
f"{'' if external_ray else 'ray stop --force; '}"
|
||||
f"{'' if external_ray else 'pkill -9 ray; '}"
|
||||
"pkill -9 slime; "
|
||||
"sleep 3; "
|
||||
f"{'' if external_ray else 'pkill -9 ray; '}"
|
||||
"pkill -9 slime; "
|
||||
"pkill -9 redis; "
|
||||
"true;"
|
||||
)
|
||||
|
||||
U.exec_command(
|
||||
f"CUDA_VISIBLE_DEVICES=6,7 python3 -m sglang.launch_server "
|
||||
f"--model-path /root/models/{TEACHER_MODEL_NAME} "
|
||||
f"--host 0.0.0.0 "
|
||||
f"--port {TEACHER_PORT} "
|
||||
f"--tp 2 "
|
||||
f"--chunked-prefill-size 4096 "
|
||||
f"--mem-fraction-static 0.6 "
|
||||
f"--context-length 32768 "
|
||||
f"> {LOG_FILE} 2>&1 & "
|
||||
)
|
||||
|
||||
U.exec_command(
|
||||
f"until curl -sf http://{TEACHER_IP}:{TEACHER_PORT}/health_generate > /dev/null; do "
|
||||
f" echo 'Waiting for teacher model...'; "
|
||||
f" tail -n 10 {LOG_FILE}; sleep 5; done; "
|
||||
f"echo 'Teacher model ready at {TEACHER_IP}:{TEACHER_PORT}.'; sleep 10;"
|
||||
)
|
||||
|
||||
|
||||
def prepare():
|
||||
U.exec_command("mkdir -p /root/models /root/datasets")
|
||||
U.exec_command(f"huggingface-cli download Qwen/{TEACHER_MODEL_NAME} --local-dir /root/models/{TEACHER_MODEL_NAME}")
|
||||
|
||||
U.convert_checkpoint(
|
||||
model_name=MODEL_NAME,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
hf_checkpoint=SFT_CHECKPOINT,
|
||||
)
|
||||
|
||||
deploy_teacher_model()
|
||||
|
||||
|
||||
def execute(rerun=True):
|
||||
load_save_path = f"/root/models/{MODEL_NAME}_ckpt__{Path(__file__).stem}/"
|
||||
|
||||
ckpt_args = (
|
||||
f"--hf-checkpoint {SFT_CHECKPOINT} "
|
||||
f"--ref-load /root/models/{MODEL_NAME}_torch_dist "
|
||||
f"--load {load_save_path} "
|
||||
f"--save {load_save_path} "
|
||||
"--save-interval 10 "
|
||||
"--save-retain-interval 10 "
|
||||
)
|
||||
|
||||
rollout_args = (
|
||||
"--prompt-data /root/datasets/dapo-math-17k/dapo-math-17k.jsonl "
|
||||
"--input-key prompt "
|
||||
"--label-key label "
|
||||
"--apply-chat-template "
|
||||
"--rollout-shuffle "
|
||||
"--num-rollout 3000 "
|
||||
"--rollout-batch-size 64 "
|
||||
"--n-samples-per-prompt 4 "
|
||||
"--rollout-max-response-len 4096 "
|
||||
"--rollout-temperature 0.8 "
|
||||
"--global-batch-size 256 "
|
||||
"--balance-data "
|
||||
)
|
||||
|
||||
rm_args = (
|
||||
"--custom-rm-path slime.rollout.on_policy_distillation.reward_func "
|
||||
"--custom-reward-post-process-path slime.rollout.on_policy_distillation.post_process_rewards "
|
||||
f"--rm-url http://{TEACHER_IP}:{TEACHER_PORT}/generate "
|
||||
"--include-verifiable-reward "
|
||||
)
|
||||
|
||||
perf_args = (
|
||||
"--tensor-model-parallel-size 2 "
|
||||
"--sequence-parallel "
|
||||
"--pipeline-model-parallel-size 1 "
|
||||
"--context-parallel-size 1 "
|
||||
"--expert-model-parallel-size 1 "
|
||||
"--expert-tensor-parallel-size 1 "
|
||||
"--recompute-granularity full "
|
||||
"--recompute-method uniform "
|
||||
"--recompute-num-layers 1 "
|
||||
"--use-dynamic-batch-size "
|
||||
"--max-tokens-per-gpu 16384 "
|
||||
)
|
||||
|
||||
grpo_args = (
|
||||
"--advantage-estimator on_policy_distillation "
|
||||
"--use-kl-loss "
|
||||
"--kl-loss-coef 0.00 "
|
||||
"--kl-loss-type low_var_kl "
|
||||
"--entropy-coef 0.00 "
|
||||
)
|
||||
|
||||
optimizer_args = (
|
||||
"--optimizer adam "
|
||||
"--lr 2e-6 "
|
||||
"--lr-decay-style constant "
|
||||
"--weight-decay 0.1 "
|
||||
"--adam-beta1 0.9 "
|
||||
"--adam-beta2 0.98 "
|
||||
)
|
||||
|
||||
wandb_args = ""
|
||||
if os.environ.get("WANDB_KEY"):
|
||||
wandb_args = (
|
||||
"--use-wandb "
|
||||
"--wandb-project lightning-opd "
|
||||
f"--wandb-group {Path(__file__).stem} "
|
||||
f"--wandb-key {os.environ['WANDB_KEY']} "
|
||||
)
|
||||
|
||||
sglang_args = (
|
||||
"--rollout-num-gpus-per-engine 1 "
|
||||
"--sglang-mem-fraction-static 0.4 "
|
||||
)
|
||||
|
||||
misc_args = (
|
||||
"--attention-dropout 0.0 "
|
||||
"--hidden-dropout 0.0 "
|
||||
"--accumulate-allreduce-grads-in-fp32 "
|
||||
"--attention-softmax-in-fp32 "
|
||||
"--attention-backend flash "
|
||||
"--actor-num-nodes 1 "
|
||||
"--actor-num-gpus-per-node 2 "
|
||||
"--rollout-num-gpus 4 "
|
||||
)
|
||||
|
||||
train_args = (
|
||||
f"{ckpt_args} "
|
||||
f"{rollout_args} "
|
||||
f"{rm_args} "
|
||||
f"{grpo_args} "
|
||||
f"{optimizer_args} "
|
||||
f"{wandb_args} "
|
||||
f"{perf_args} "
|
||||
f"{sglang_args} "
|
||||
f"{misc_args} "
|
||||
)
|
||||
|
||||
U.execute_train(
|
||||
rerun=rerun,
|
||||
train_args=train_args,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
prepare()
|
||||
execute(rerun=False)
|
||||
190
configs/opd/qwen3-8b-opd.py
Normal file
190
configs/opd/qwen3-8b-opd.py
Normal file
@@ -0,0 +1,190 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import slime.utils.external_utils.command_utils as U
|
||||
from slime.utils.external_utils.command_utils import get_bool_env_var
|
||||
|
||||
# Standard OPD: requires a live teacher server during training.
|
||||
# 4 GPUs for actor (TP=4), 2 GPUs for rollout, 2 GPUs for teacher server.
|
||||
#
|
||||
# Required env vars:
|
||||
# SFT_CHECKPOINT - path to the SFT checkpoint (HF format)
|
||||
|
||||
MODEL_NAME = "Qwen3-8B-Base-Open-Thoughts-Qwen3-32B-sft-3.5k"
|
||||
TEACHER_MODEL_NAME = "Qwen3-32B"
|
||||
TEACHER_IP = os.environ.get("MASTER_ADDR", "127.0.0.1")
|
||||
TEACHER_PORT = 13141
|
||||
MODEL_TYPE = "qwen3-8B"
|
||||
NUM_GPUS = 8
|
||||
SFT_CHECKPOINT = os.environ["SFT_CHECKPOINT"]
|
||||
|
||||
|
||||
def deploy_teacher_model():
|
||||
import random, string
|
||||
random_suffix = ''.join(random.choices(string.ascii_letters + string.digits, k=6))
|
||||
LOG_FILE = f"/tmp/sglang_{random_suffix}.log"
|
||||
external_ray = get_bool_env_var("SLIME_SCRIPT_EXTERNAL_RAY")
|
||||
|
||||
U.exec_command(
|
||||
"pkill -9 sglang; "
|
||||
"sleep 3; "
|
||||
f"{'' if external_ray else 'ray stop --force; '}"
|
||||
f"{'' if external_ray else 'pkill -9 ray; '}"
|
||||
"pkill -9 slime; "
|
||||
"sleep 3; "
|
||||
f"{'' if external_ray else 'pkill -9 ray; '}"
|
||||
"pkill -9 slime; "
|
||||
"pkill -9 redis; "
|
||||
"true;"
|
||||
)
|
||||
|
||||
U.exec_command(
|
||||
f"CUDA_VISIBLE_DEVICES=6,7 python3 -m sglang.launch_server "
|
||||
f"--model-path /root/models/{TEACHER_MODEL_NAME} "
|
||||
f"--host 0.0.0.0 "
|
||||
f"--port {TEACHER_PORT} "
|
||||
f"--tp 2 "
|
||||
f"--chunked-prefill-size 4096 "
|
||||
f"--mem-fraction-static 0.6 "
|
||||
f"--context-length 32768 "
|
||||
f"> {LOG_FILE} 2>&1 & "
|
||||
)
|
||||
|
||||
U.exec_command(
|
||||
f"until curl -sf http://{TEACHER_IP}:{TEACHER_PORT}/health_generate > /dev/null; do "
|
||||
f" echo 'Waiting for teacher model...'; "
|
||||
f" tail -n 10 {LOG_FILE}; sleep 5; done; "
|
||||
f"echo 'Teacher model ready at {TEACHER_IP}:{TEACHER_PORT}.'; sleep 10;"
|
||||
)
|
||||
|
||||
|
||||
def prepare():
|
||||
U.exec_command("mkdir -p /root/models /root/datasets/dapo-math-17k")
|
||||
U.exec_command(f"huggingface-cli download Qwen/{TEACHER_MODEL_NAME} --local-dir /root/models/{TEACHER_MODEL_NAME}")
|
||||
|
||||
U.convert_checkpoint(
|
||||
model_name=MODEL_NAME,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
hf_checkpoint=SFT_CHECKPOINT,
|
||||
)
|
||||
|
||||
deploy_teacher_model()
|
||||
|
||||
|
||||
def execute(rerun=True):
|
||||
load_save_path = f"/root/models/{MODEL_NAME}_ckpt__{Path(__file__).stem}/"
|
||||
|
||||
ckpt_args = (
|
||||
f"--hf-checkpoint {SFT_CHECKPOINT} "
|
||||
f"--ref-load /root/models/{MODEL_NAME}_torch_dist "
|
||||
f"--load {load_save_path} "
|
||||
f"--save {load_save_path} "
|
||||
"--save-interval 5 "
|
||||
"--save-retain-interval 5 "
|
||||
)
|
||||
|
||||
rollout_args = (
|
||||
"--prompt-data /root/datasets/dapo-math-17k/dapo-math-17k.jsonl "
|
||||
"--input-key prompt "
|
||||
"--label-key label "
|
||||
"--apply-chat-template "
|
||||
"--rollout-shuffle "
|
||||
"--num-rollout 3000 "
|
||||
"--rollout-batch-size 64 "
|
||||
"--n-samples-per-prompt 4 "
|
||||
"--rollout-max-response-len 4096 "
|
||||
"--rollout-temperature 0.8 "
|
||||
"--global-batch-size 256 "
|
||||
"--balance-data "
|
||||
)
|
||||
|
||||
rm_args = (
|
||||
"--custom-rm-path slime.rollout.on_policy_distillation.reward_func "
|
||||
"--custom-reward-post-process-path slime.rollout.on_policy_distillation.post_process_rewards "
|
||||
f"--rm-url http://{TEACHER_IP}:{TEACHER_PORT}/generate "
|
||||
"--include-verifiable-reward "
|
||||
)
|
||||
|
||||
perf_args = (
|
||||
"--tensor-model-parallel-size 4 "
|
||||
"--sequence-parallel "
|
||||
"--pipeline-model-parallel-size 1 "
|
||||
"--context-parallel-size 1 "
|
||||
"--expert-model-parallel-size 1 "
|
||||
"--expert-tensor-parallel-size 1 "
|
||||
"--recompute-granularity full "
|
||||
"--recompute-method uniform "
|
||||
"--recompute-num-layers 1 "
|
||||
"--use-dynamic-batch-size "
|
||||
"--max-tokens-per-gpu 8192 "
|
||||
)
|
||||
|
||||
grpo_args = (
|
||||
"--advantage-estimator on_policy_distillation "
|
||||
"--use-kl-loss "
|
||||
"--kl-loss-coef 0.00 "
|
||||
"--kl-loss-type low_var_kl "
|
||||
"--entropy-coef 0.00 "
|
||||
)
|
||||
|
||||
optimizer_args = (
|
||||
"--optimizer adam "
|
||||
"--lr 2e-6 "
|
||||
"--lr-decay-style constant "
|
||||
"--weight-decay 0.1 "
|
||||
"--adam-beta1 0.9 "
|
||||
"--adam-beta2 0.98 "
|
||||
)
|
||||
|
||||
wandb_args = ""
|
||||
if os.environ.get("WANDB_KEY"):
|
||||
wandb_args = (
|
||||
"--use-wandb "
|
||||
"--wandb-project lightning-opd "
|
||||
f"--wandb-group {Path(__file__).stem} "
|
||||
f"--wandb-key {os.environ['WANDB_KEY']} "
|
||||
)
|
||||
|
||||
sglang_args = (
|
||||
"--rollout-num-gpus-per-engine 1 "
|
||||
"--sglang-mem-fraction-static 0.4 "
|
||||
)
|
||||
|
||||
misc_args = (
|
||||
"--attention-dropout 0.0 "
|
||||
"--hidden-dropout 0.0 "
|
||||
"--accumulate-allreduce-grads-in-fp32 "
|
||||
"--attention-softmax-in-fp32 "
|
||||
"--attention-backend flash "
|
||||
"--actor-num-nodes 1 "
|
||||
"--actor-num-gpus-per-node 4 "
|
||||
"--rollout-num-gpus 2 "
|
||||
)
|
||||
|
||||
train_args = (
|
||||
f"{ckpt_args} "
|
||||
f"{rollout_args} "
|
||||
f"{rm_args} "
|
||||
f"{grpo_args} "
|
||||
f"{optimizer_args} "
|
||||
f"{wandb_args} "
|
||||
f"{perf_args} "
|
||||
f"{sglang_args} "
|
||||
f"{misc_args} "
|
||||
)
|
||||
|
||||
U.execute_train(
|
||||
rerun=rerun,
|
||||
train_args=train_args,
|
||||
num_gpus_per_node=NUM_GPUS,
|
||||
megatron_model_type=MODEL_TYPE,
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
prepare()
|
||||
execute(rerun=False)
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:66cf6cffd11dd523d28cd449afcd5500c62ab47e6617a977039e931d22ae2689
|
||||
size 13944494336
|
||||
23
configs/sft/dataset_info.json
Normal file
23
configs/sft/dataset_info.json
Normal file
@@ -0,0 +1,23 @@
|
||||
{
|
||||
"openthoughts3_300k_qwen3-8b": {
|
||||
"file_name": "data/sft_data/openthoughts3_300k_qwen3-8b.jsonl",
|
||||
"formatting": "sharegpt",
|
||||
"columns": {
|
||||
"messages": "messages"
|
||||
},
|
||||
"tags": {
|
||||
"role_tag": "role",
|
||||
"content_tag": "content",
|
||||
"user_tag": "user",
|
||||
"assistant_tag": "assistant",
|
||||
"system_tag": "system"
|
||||
}
|
||||
},
|
||||
"openthoughts3_300k_qwen3-32b": {
|
||||
"file_name": "data/sft_data/openthoughts3_300k_qwen3-32b.parquet",
|
||||
"formatting": "sharegpt",
|
||||
"columns": {
|
||||
"messages": "messages"
|
||||
}
|
||||
}
|
||||
}
|
||||
16
configs/sft/dataset_info_ori_Fri_5_Jun.json
Normal file
16
configs/sft/dataset_info_ori_Fri_5_Jun.json
Normal file
@@ -0,0 +1,16 @@
|
||||
{
|
||||
"openthoughts3_300k_qwen3-8b": {
|
||||
"file_name": "data/sft_data/openthoughts3_300k_qwen3-8b.jsonl",
|
||||
"formatting": "sharegpt",
|
||||
"columns": {
|
||||
"messages": "messages"
|
||||
}
|
||||
},
|
||||
"openthoughts3_300k_qwen3-32b": {
|
||||
"file_name": "data/sft_data/openthoughts3_300k_qwen3-32b.parquet",
|
||||
"formatting": "sharegpt",
|
||||
"columns": {
|
||||
"messages": "messages"
|
||||
}
|
||||
}
|
||||
}
|
||||
42
configs/sft/qwen3-4b-base-open-thoughts3-qwen3-8b.yaml
Normal file
42
configs/sft/qwen3-4b-base-open-thoughts3-qwen3-8b.yaml
Normal file
@@ -0,0 +1,42 @@
|
||||
### model
|
||||
model_name_or_path: model_weights/qwen3-4b-base
|
||||
|
||||
### method
|
||||
stage: sft
|
||||
do_train: true
|
||||
finetuning_type: full
|
||||
deepspeed: examples/deepspeed/ds_z0_config.json
|
||||
enable_liger_kernel: true
|
||||
packing: true
|
||||
|
||||
|
||||
### dataset
|
||||
dataset: openthoughts3_300k_qwen3-8b
|
||||
template: qwen3
|
||||
cutoff_len: 16384
|
||||
overwrite_cache: true
|
||||
preprocessing_num_workers: 16
|
||||
dataloader_persistent_workers: true
|
||||
dataloader_pin_memory: true
|
||||
dataloader_num_workers: 4
|
||||
|
||||
### output
|
||||
logging_steps: 1
|
||||
save_steps: 100
|
||||
save_total_limit: 10
|
||||
plot_loss: true
|
||||
overwrite_output_dir: false
|
||||
save_only_model: false
|
||||
#report_to: wandb
|
||||
report_to: none
|
||||
run_name: qwen3-4b-base-open-thoughts3-qwen3-8b
|
||||
|
||||
### train
|
||||
per_device_train_batch_size: 4
|
||||
gradient_accumulation_steps: 2
|
||||
learning_rate: 0.00008
|
||||
max_steps: 3000
|
||||
lr_scheduler_type: cosine
|
||||
warmup_ratio: 0.1
|
||||
bf16: true
|
||||
ddp_timeout: 180000000
|
||||
41
configs/sft/qwen3-8b-base-open-thoughts3-qwen3-32b.yaml
Normal file
41
configs/sft/qwen3-8b-base-open-thoughts3-qwen3-32b.yaml
Normal file
@@ -0,0 +1,41 @@
|
||||
### model
|
||||
model_name_or_path: Qwen/Qwen3-8B-Base
|
||||
|
||||
### method
|
||||
stage: sft
|
||||
do_train: true
|
||||
finetuning_type: full
|
||||
deepspeed: examples/deepspeed/ds_z1_config.json
|
||||
enable_liger_kernel: true
|
||||
packing: true
|
||||
|
||||
|
||||
### dataset
|
||||
dataset: openthoughts3_300k_qwen3-32b
|
||||
template: qwen3
|
||||
cutoff_len: 16384
|
||||
overwrite_cache: true
|
||||
preprocessing_num_workers: 16
|
||||
dataloader_persistent_workers: true
|
||||
dataloader_pin_memory: true
|
||||
dataloader_num_workers: 4
|
||||
|
||||
### output
|
||||
logging_steps: 1
|
||||
save_steps: 100
|
||||
save_total_limit: 10
|
||||
plot_loss: true
|
||||
overwrite_output_dir: false
|
||||
save_only_model: false
|
||||
report_to: wandb
|
||||
run_name: qwen3-8b-base-open-thoughts3-qwen3-32b
|
||||
|
||||
### train
|
||||
per_device_train_batch_size: 2
|
||||
gradient_accumulation_steps: 2
|
||||
learning_rate: 8e-5
|
||||
max_steps: 3000
|
||||
lr_scheduler_type: cosine
|
||||
warmup_ratio: 0.1
|
||||
bf16: true
|
||||
ddp_timeout: 180000000
|
||||
49
configs/sft/run_sft.sh
Normal file
49
configs/sft/run_sft.sh
Normal file
@@ -0,0 +1,49 @@
|
||||
#!/usr/bin/env bash
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
# Step 2: Run SFT training with LlamaFactory.
|
||||
#
|
||||
# Required environment variables:
|
||||
# CONFIG_YAML - Name of the SFT config file in configs/sft/ (e.g. qwen3-4b-base-sft-qwen3-8b.yaml)
|
||||
# OUTPUT_DIR - Directory for the SFT checkpoint output
|
||||
#
|
||||
# Optional:
|
||||
# NUM_NODES - Number of nodes (default: 4)
|
||||
# NUM_GPUS - GPUs per node (default: 8)
|
||||
# MASTER_ADDR - Master node address (default: localhost)
|
||||
#
|
||||
# Prerequisites:
|
||||
# - LlamaFactory installed (pip install llamafactory)
|
||||
# - SFT data generated by Step 1 and registered in LlamaFactory's dataset_info.json
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
: "${CONFIG_YAML:?Set CONFIG_YAML (e.g. qwen3-4b-base-sft-qwen3-8b.yaml)}"
|
||||
: "${OUTPUT_DIR:?Set OUTPUT_DIR for SFT checkpoint output}"
|
||||
|
||||
NUM_NODES="${NUM_NODES:-4}"
|
||||
NUM_GPUS="${NUM_GPUS:-8}"
|
||||
MASTER_ADDR="${MASTER_ADDR:-localhost}"
|
||||
MASTER_PORT="${MASTER_PORT:-29500}"
|
||||
|
||||
# torchrun \
|
||||
# --nnodes "${NUM_NODES}" \
|
||||
# --nproc_per_node="${NUM_GPUS}" \
|
||||
# --rdzv_id $RANDOM \
|
||||
# --rdzv_backend c10d \
|
||||
# --rdzv_endpoint "${MASTER_ADDR}:29500" \
|
||||
# -m llamafactory.cli.train \
|
||||
# "configs/sft/${CONFIG_YAML}" \
|
||||
# "dataset_dir=configs/sft" \
|
||||
# "output_dir=${OUTPUT_DIR}"
|
||||
|
||||
FORCE_TORCHRUN=1 \
|
||||
NNODES="${NUM_NODES}" \
|
||||
NPROC_PER_NODE="${NUM_GPUS}" \
|
||||
MASTER_ADDR="${MASTER_ADDR}" \
|
||||
MASTER_PORT="${MASTER_PORT}" \
|
||||
llamafactory-cli train \
|
||||
"configs/sft/${CONFIG_YAML}" \
|
||||
"dataset_dir=configs/sft" \
|
||||
"output_dir=${OUTPUT_DIR}"
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:546f8126bc491dfdf620a3a670044f64506d4e5f943d4c29e6aa30c59e5a4d68
|
||||
size 480095067
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:1d5d698fde86339a0d56ad7f9676fbeee81016ca36c5788569cfaab4ba12eb01
|
||||
size 179933493
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:443e972c427905ccbcac9391a5caf701c895187838f1f25dbc922f090b5b8c3a
|
||||
size 431472722
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:ef6fdd7bd424e5a9f8ddd2ba1c212705b1c4fd6cf3ec08475d0b3eb070ffba62
|
||||
size 791187240
|
||||
3
data/prompts/dapo-math-17k/dapo-math-17k.jsonl
Normal file
3
data/prompts/dapo-math-17k/dapo-math-17k.jsonl
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cc9c39c2aa19177abe9464741e121cf4cac90fd25484ef3cdf86535101e3a5b6
|
||||
size 10490834
|
||||
3
data/prompts/openthoughts3_300k.jsonl
Normal file
3
data/prompts/openthoughts3_300k.jsonl
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:074b751c84563e6251891015474f60f44ded7770baea53bd6511ba6dc7625f83
|
||||
size 250211763
|
||||
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:cc101946ded7ebb333d45a38ac98545366fe049423263d2a09de093ffa19730a
|
||||
size 194935550
|
||||
160
data_curation/merge.py
Normal file
160
data_curation/merge.py
Normal file
@@ -0,0 +1,160 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
"""
|
||||
Merge Arrow IPC files produced by data_curation/pipeline.py into a single parquet.
|
||||
|
||||
After multi-GPU data generation, each worker writes Arrow files into
|
||||
rank-specific subdirectories. This script merges them into one parquet
|
||||
file for downstream consumption (SFT training or Lightning OPD preparation).
|
||||
|
||||
Usage:
|
||||
python data_curation/merge.py \
|
||||
--input-dir data/sft_data \
|
||||
--output data/sft_data/merged.parquet
|
||||
|
||||
# With filtering: only keep samples with token count <= 16384
|
||||
python data_curation/merge.py \
|
||||
--input-dir data/sft_data \
|
||||
--output data/sft_data/merged.parquet \
|
||||
--max-tokens 16384
|
||||
"""
|
||||
|
||||
import argparse
|
||||
from pathlib import Path
|
||||
|
||||
import json
|
||||
import pyarrow as pa
|
||||
import pyarrow.ipc as ipc
|
||||
from tqdm import tqdm
|
||||
|
||||
|
||||
def parse_args():
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Merge Arrow IPC files into a single parquet file."
|
||||
)
|
||||
parser.add_argument(
|
||||
"--input-dir", type=str, required=True,
|
||||
help="Directory containing Arrow files (searched recursively).",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output", type=str, required=True,
|
||||
help="Output parquet file path.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--max-tokens", type=int, default=None,
|
||||
help="If set, discard rows with tokens > this value.",
|
||||
)
|
||||
return parser.parse_args()
|
||||
|
||||
|
||||
# def merge_arrow_files(input_dir: str, output: str, max_tokens: int | None = None):
|
||||
# input_path = Path(input_dir)
|
||||
# arrow_files = sorted(input_path.rglob("*.arrow"))
|
||||
|
||||
# if not arrow_files:
|
||||
# print(f"No Arrow files found in {input_dir}")
|
||||
# return
|
||||
|
||||
# print(f"Found {len(arrow_files)} Arrow files in {input_dir}")
|
||||
|
||||
# tables = []
|
||||
# total_rows = 0
|
||||
# for f in tqdm(arrow_files, desc="Reading Arrow files"):
|
||||
# with pa.OSFile(str(f), "rb") as source:
|
||||
# table = ipc.open_file(source).read_all()
|
||||
# tables.append(table)
|
||||
# total_rows += len(table)
|
||||
|
||||
# merged = pa.concat_tables(tables)
|
||||
# print(f"Total rows before filtering: {total_rows}")
|
||||
|
||||
# if max_tokens is not None and "tokens" in merged.column_names:
|
||||
# tokens = merged.column("tokens").to_pylist()
|
||||
# mask = [t <= max_tokens for t in tokens]
|
||||
# merged = merged.filter(mask)
|
||||
# filtered = total_rows - len(merged)
|
||||
# print(f"Filtered {filtered} rows with tokens > {max_tokens}")
|
||||
|
||||
# output_path = Path(output)
|
||||
# output_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
# df = merged.to_pandas()
|
||||
# df.to_parquet(output, index=False)
|
||||
|
||||
# print(f"Merged {len(df)} rows -> {output}")
|
||||
|
||||
def merge_arrow_files(input_dir: str, output: str, max_tokens: int | None = None):
|
||||
input_path = Path(input_dir)
|
||||
arrow_files = sorted(input_path.rglob("*.arrow"))
|
||||
|
||||
if not arrow_files:
|
||||
print(f"No Arrow files found in {input_dir}")
|
||||
return
|
||||
|
||||
print(f"Found {len(arrow_files)} Arrow files in {input_dir}")
|
||||
|
||||
output_path = Path(output)
|
||||
output_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
# Recommended path for nested conversation data
|
||||
if output_path.suffix == ".jsonl":
|
||||
total_rows = 0
|
||||
kept_rows = 0
|
||||
filtered_rows = 0
|
||||
|
||||
with open(output_path, "w", encoding="utf-8") as out_f:
|
||||
for f in tqdm(arrow_files, desc="Reading Arrow files"):
|
||||
with pa.OSFile(str(f), "rb") as source:
|
||||
table = ipc.open_file(source).read_all()
|
||||
|
||||
rows = table.to_pylist()
|
||||
total_rows += len(rows)
|
||||
|
||||
for row in rows:
|
||||
if max_tokens is not None and "tokens" in row:
|
||||
if row["tokens"] > max_tokens:
|
||||
filtered_rows += 1
|
||||
continue
|
||||
|
||||
out_f.write(json.dumps(row, ensure_ascii=False) + "\n")
|
||||
kept_rows += 1
|
||||
|
||||
print(f"Total rows before filtering: {total_rows}")
|
||||
if max_tokens is not None:
|
||||
print(f"Filtered {filtered_rows} rows with tokens > {max_tokens}")
|
||||
print(f"Merged {kept_rows} rows -> {output}")
|
||||
return
|
||||
|
||||
# Optional parquet path, not recommended for nested messages
|
||||
elif output_path.suffix == ".parquet":
|
||||
import pyarrow.parquet as pq
|
||||
|
||||
tables = []
|
||||
total_rows = 0
|
||||
|
||||
for f in tqdm(arrow_files, desc="Reading Arrow files"):
|
||||
with pa.OSFile(str(f), "rb") as source:
|
||||
table = ipc.open_file(source).read_all()
|
||||
tables.append(table)
|
||||
total_rows += len(table)
|
||||
|
||||
merged = pa.concat_tables(tables)
|
||||
print(f"Total rows before filtering: {total_rows}")
|
||||
|
||||
if max_tokens is not None and "tokens" in merged.column_names:
|
||||
tokens = merged.column("tokens").to_pylist()
|
||||
mask = [t <= max_tokens for t in tokens]
|
||||
merged = merged.filter(mask)
|
||||
filtered = total_rows - len(merged)
|
||||
print(f"Filtered {filtered} rows with tokens > {max_tokens}")
|
||||
|
||||
pq.write_table(merged, output)
|
||||
print(f"Merged {len(merged)} rows -> {output}")
|
||||
return
|
||||
|
||||
else:
|
||||
raise ValueError(f"Unsupported output format: {output_path.suffix}")
|
||||
|
||||
if __name__ == "__main__":
|
||||
args = parse_args()
|
||||
merge_arrow_files(args.input_dir, args.output, args.max_tokens)
|
||||
218
data_curation/pipeline.py
Normal file
218
data_curation/pipeline.py
Normal file
@@ -0,0 +1,218 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
"""
|
||||
Data curation pipeline: generate responses from a dataset using vLLM.
|
||||
|
||||
Each worker (identified by --rank) processes a disjoint shard of the input
|
||||
dataset, generates responses via vLLM offline inference, and writes results
|
||||
as Arrow IPC files (one per batch) into a rank-specific output directory.
|
||||
Checkpointing allows resuming from the last completed batch.
|
||||
|
||||
Standalone:
|
||||
python data_curation/pipeline.py \
|
||||
--model Qwen/Qwen3-4B \
|
||||
--input data.jsonl \
|
||||
--output-dir output/
|
||||
|
||||
Multi-GPU (one model per GPU):
|
||||
See run_curation.sh for the recommended launch pattern.
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import os
|
||||
import pickle
|
||||
from pathlib import Path
|
||||
|
||||
import pandas as pd
|
||||
import pyarrow as pa
|
||||
import pyarrow.ipc as ipc
|
||||
from tqdm import tqdm
|
||||
from vllm import LLM, SamplingParams
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Data I/O
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
def load_dataset(path: str) -> list[dict]:
|
||||
"""Load a .jsonl or .parquet dataset into a list of dicts."""
|
||||
if path.endswith(".parquet"):
|
||||
df = pd.read_parquet(path)
|
||||
records = df.to_dict("records")
|
||||
for record in records:
|
||||
if "prompt" in record and hasattr(record["prompt"], "tolist"):
|
||||
record["prompt"] = record["prompt"].tolist()
|
||||
return records
|
||||
elif path.endswith(".jsonl"):
|
||||
with open(path) as f:
|
||||
return [json.loads(line) for line in f]
|
||||
else:
|
||||
raise ValueError(f"Unsupported format: {path}. Use .jsonl or .parquet.")
|
||||
|
||||
|
||||
def save_batch_arrow(rows: list[dict], path: str) -> None:
|
||||
"""Write a list of dicts as an Arrow IPC file."""
|
||||
table = pa.Table.from_pandas(pd.DataFrame(rows))
|
||||
with pa.OSFile(path, "wb") as sink:
|
||||
with ipc.new_file(sink, table.schema) as writer:
|
||||
writer.write_table(table)
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Core pipeline
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
def run_curation(args: argparse.Namespace) -> None:
|
||||
tag = f"[Rank {args.rank}/{args.world_size}]"
|
||||
|
||||
# ── Load & shard dataset ──────────────────────────────────────────────
|
||||
print(f"{tag} Loading dataset: {args.input}")
|
||||
dataset = load_dataset(args.input)
|
||||
|
||||
if args.num_samples is not None:
|
||||
dataset = dataset[: args.num_samples]
|
||||
print(f"{tag} Debug mode: limiting to {args.num_samples} samples")
|
||||
|
||||
if args.world_size > 1:
|
||||
dataset = dataset[args.rank :: args.world_size]
|
||||
print(f"{tag} Assigned {len(dataset)} samples")
|
||||
|
||||
# ── Output directory ──────────────────────────────────────────────────
|
||||
if args.world_size > 1:
|
||||
output_dir = Path(args.output_dir) / f"rank{args.rank:05d}"
|
||||
else:
|
||||
output_dir = Path(args.output_dir)
|
||||
output_dir.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
# ── Checkpoint ────────────────────────────────────────────────────────
|
||||
ckpt_dir = Path(args.checkpoint_dir)
|
||||
ckpt_dir.mkdir(parents=True, exist_ok=True)
|
||||
ckpt_file = ckpt_dir / f"rank{args.rank:05d}.pkl"
|
||||
|
||||
start_idx = 0
|
||||
if ckpt_file.exists():
|
||||
with open(ckpt_file, "rb") as f:
|
||||
start_idx = pickle.load(f)["next_idx"]
|
||||
print(f"{tag} Resuming from index {start_idx}")
|
||||
|
||||
# ── Model ─────────────────────────────────────────────────────────────
|
||||
print(f"{tag} Loading model: {args.model} (tp={args.tensor_parallel_size})")
|
||||
llm = LLM(
|
||||
model=args.model,
|
||||
tensor_parallel_size=args.tensor_parallel_size,
|
||||
trust_remote_code=True,
|
||||
)
|
||||
|
||||
sampling_params = SamplingParams(
|
||||
temperature=args.temperature,
|
||||
top_p=args.top_p,
|
||||
max_tokens=args.max_tokens,
|
||||
n=args.num_responses,
|
||||
)
|
||||
|
||||
# ── Batch loop ────────────────────────────────────────────────────────
|
||||
total_batches = (len(dataset) + args.batch_size - 1) // args.batch_size
|
||||
total_saved = 0
|
||||
|
||||
print(f"{tag} Processing {len(dataset)} prompts, batch_size={args.batch_size}, "
|
||||
f"total_batches={total_batches}")
|
||||
|
||||
for batch_start in range(start_idx, len(dataset), args.batch_size):
|
||||
batch_end = min(batch_start + args.batch_size, len(dataset))
|
||||
batch = dataset[batch_start:batch_end]
|
||||
batch_idx = batch_start // args.batch_size
|
||||
|
||||
prompts = [item["prompt"] for item in batch]
|
||||
print(f"{tag} Batch {batch_idx + 1}/{total_batches} "
|
||||
f"({batch_end - batch_start} samples) ...")
|
||||
|
||||
outputs = llm.chat(prompts, sampling_params)
|
||||
|
||||
# Build results
|
||||
rows = []
|
||||
for item, output in zip(batch, outputs):
|
||||
for completion in output.outputs:
|
||||
text = completion.text
|
||||
# Ensure <think> tag is present
|
||||
if "</think>" in text and not text.strip().startswith("<think>"):
|
||||
text = "<think>\n" + text
|
||||
messages = item["prompt"] + [{"role": "assistant", "content": text}]
|
||||
rows.append({
|
||||
"messages": messages,
|
||||
"tokens": len(completion.token_ids),
|
||||
})
|
||||
|
||||
# Save Arrow file
|
||||
arrow_path = output_dir / f"data-{batch_idx:05d}-of-{total_batches:05d}.arrow"
|
||||
save_batch_arrow(rows, str(arrow_path))
|
||||
total_saved += len(rows)
|
||||
|
||||
# Save checkpoint
|
||||
with open(ckpt_file, "wb") as f:
|
||||
pickle.dump({"next_idx": batch_end}, f)
|
||||
|
||||
print(f"{tag} Saved {arrow_path.name} (total: {total_saved})")
|
||||
|
||||
# ── Cleanup ───────────────────────────────────────────────────────────
|
||||
if ckpt_file.exists():
|
||||
ckpt_file.unlink()
|
||||
print(f"{tag} Done! {total_saved} samples → {output_dir}/")
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# CLI
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
def parse_args() -> argparse.Namespace:
|
||||
p = argparse.ArgumentParser(
|
||||
description="Generate responses from a dataset using vLLM offline inference.",
|
||||
)
|
||||
# Required
|
||||
p.add_argument("--model", type=str, required=True,
|
||||
help="HuggingFace model name or path.")
|
||||
p.add_argument("--input", type=str, required=True,
|
||||
help="Input dataset (.jsonl or .parquet).")
|
||||
p.add_argument("--output-dir", type=str, required=True,
|
||||
help="Root output directory. Each rank writes to a subdirectory.")
|
||||
|
||||
# Generation
|
||||
p.add_argument("--max-tokens", type=int, default=16384,
|
||||
help="Max new tokens per response (default: 16384).")
|
||||
p.add_argument("--temperature", type=float, default=0.7,
|
||||
help="Sampling temperature (default: 0.7).")
|
||||
p.add_argument("--top-p", type=float, default=0.9,
|
||||
help="Nucleus sampling top-p (default: 0.9).")
|
||||
p.add_argument("--num-responses", type=int, default=1,
|
||||
help="Number of responses per prompt (default: 1).")
|
||||
p.add_argument("--batch-size", type=int, default=32,
|
||||
help="Prompts per vLLM batch call (default: 32).")
|
||||
|
||||
# Parallelism
|
||||
p.add_argument("--tensor-parallel-size", type=int, default=1,
|
||||
help="vLLM tensor-parallel size (default: 1).")
|
||||
p.add_argument("--rank", type=int, default=None,
|
||||
help="Worker rank (auto-detected from env if omitted).")
|
||||
p.add_argument("--world-size", type=int, default=None,
|
||||
help="Total workers (auto-detected from env if omitted).")
|
||||
|
||||
# Misc
|
||||
p.add_argument("--num-samples", type=int, default=None,
|
||||
help="Limit total samples before sharding (for debugging).")
|
||||
p.add_argument("--checkpoint-dir", type=str, default="checkpoints",
|
||||
help="Directory for per-rank checkpoint files (default: checkpoints).")
|
||||
|
||||
args = p.parse_args()
|
||||
|
||||
# Auto-detect rank / world_size from environment (torchrun, etc.)
|
||||
if args.rank is None:
|
||||
args.rank = int(os.environ.get("RANK", os.environ.get("LOCAL_RANK", 0)))
|
||||
if args.world_size is None:
|
||||
args.world_size = int(os.environ.get("WORLD_SIZE", 1))
|
||||
|
||||
return args
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
run_curation(parse_args())
|
||||
247
data_curation/prepare_lightning_opd.py
Normal file
247
data_curation/prepare_lightning_opd.py
Normal file
@@ -0,0 +1,247 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
"""
|
||||
Prepare Lightning OPD parquet from student rollout data.
|
||||
|
||||
Phase 1 – tokenize (CPU-friendly):
|
||||
Reads student rollout parquet, builds prompt via chat template,
|
||||
tokenizes responses, truncates to --max-response-len, writes intermediate
|
||||
parquet WITHOUT teacher logprobs.
|
||||
|
||||
Phase 2 – precompute teacher logprobs (requires GPU / teacher sglang server):
|
||||
Reads the intermediate parquet produced in Phase 1, sends each
|
||||
(prompt + response) sequence to the teacher sglang server, stores
|
||||
per-token response logprobs back into the metadata, writes the final
|
||||
parquet.
|
||||
|
||||
Usage (Phase 1, CPU node):
|
||||
python3 data_curation/prepare_lightning_opd.py \\
|
||||
--tokenizer-path checkpoints/sft \\
|
||||
--input-parquet data/rollouts/rollouts.parquet \\
|
||||
--output-dir data/lightning_opd
|
||||
|
||||
Usage (Phase 2, GPU node with teacher sglang running):
|
||||
python3 data_curation/prepare_lightning_opd.py \\
|
||||
--tokenizer-path checkpoints/sft \\
|
||||
--input-parquet data/rollouts/rollouts.parquet \\
|
||||
--output-dir data/lightning_opd \\
|
||||
--compute-teacher-logprobs \\
|
||||
--teacher-url http://127.0.0.1:13141/generate
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import asyncio
|
||||
from pathlib import Path
|
||||
|
||||
import aiohttp
|
||||
import pandas as pd
|
||||
from transformers import AutoTokenizer
|
||||
from tqdm import tqdm
|
||||
|
||||
|
||||
def parse_args():
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Prepare Lightning OPD parquet data (tokenize + optional teacher logprobs)."
|
||||
)
|
||||
parser.add_argument(
|
||||
"--tokenizer-path", type=str, required=True,
|
||||
help="Path to HuggingFace tokenizer (e.g. the student SFT checkpoint).",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--input-parquet", type=str, required=True,
|
||||
help="Path to student rollout parquet. Expected columns: messages (list[dict]), tokens (int).",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output-dir", type=str, required=True,
|
||||
help="Directory where intermediate and final parquet files are written.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--max-response-len", type=int, default=4096,
|
||||
help="Maximum response token length; longer responses are truncated (default: 4096).",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--compute-teacher-logprobs", action="store_true",
|
||||
help="Run Phase 2: compute teacher logprobs via a running sglang server.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--teacher-url", type=str, default="http://127.0.0.1:13141/generate",
|
||||
help="Teacher sglang server URL (default: http://127.0.0.1:13141/generate).",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--concurrency", type=int, default=64,
|
||||
help="Number of concurrent requests to teacher sglang server (default: 64).",
|
||||
)
|
||||
return parser.parse_args()
|
||||
|
||||
|
||||
# ── Phase 1: tokenize ────────────────────────────────────────────────────────
|
||||
|
||||
def phase1_tokenize(args, intermediate_path: Path):
|
||||
print(f"[Phase 1] Loading tokenizer from {args.tokenizer_path}")
|
||||
tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_path, trust_remote_code=True)
|
||||
|
||||
print(f"[Phase 1] Loading input parquet: {args.input_parquet}")
|
||||
df = pd.read_parquet(args.input_parquet)
|
||||
print(f"[Phase 1] Total rows: {len(df)}")
|
||||
|
||||
rows_out = []
|
||||
truncated = 0
|
||||
skipped = 0
|
||||
|
||||
for row in tqdm(df.itertuples(), total=len(df), desc="Tokenizing"):
|
||||
messages = row.messages
|
||||
|
||||
user_messages = [m for m in messages if m["role"] != "assistant"]
|
||||
prompt_str = tokenizer.apply_chat_template(
|
||||
user_messages, tokenize=False, add_generation_prompt=True, enable_thinking=True
|
||||
)
|
||||
|
||||
assistant_msg = None
|
||||
for msg in messages:
|
||||
if msg["role"] == "assistant":
|
||||
assistant_msg = msg["content"]
|
||||
break
|
||||
if assistant_msg is None:
|
||||
skipped += 1
|
||||
continue
|
||||
|
||||
response_ids = tokenizer.encode(assistant_msg, add_special_tokens=False)
|
||||
|
||||
if len(response_ids) > args.max_response_len:
|
||||
truncated += 1
|
||||
response_ids = response_ids[:args.max_response_len]
|
||||
assistant_msg = tokenizer.decode(response_ids, skip_special_tokens=False)
|
||||
|
||||
rows_out.append({
|
||||
"prompt": prompt_str,
|
||||
"label": "0",
|
||||
"metadata": {
|
||||
"is_lightning_opd": True,
|
||||
"response_tokens": response_ids,
|
||||
"loss_mask": [1] * len(response_ids),
|
||||
"response": assistant_msg,
|
||||
},
|
||||
})
|
||||
|
||||
print(f"[Phase 1] Rows written: {len(rows_out)}, "
|
||||
f"truncated to {args.max_response_len}: {truncated}, skipped: {skipped}")
|
||||
df_out = pd.DataFrame(rows_out)
|
||||
intermediate_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
df_out.to_parquet(intermediate_path, index=False)
|
||||
print(f"[Phase 1] Saved to {intermediate_path}")
|
||||
|
||||
|
||||
# ── Phase 2: precompute teacher logprobs ─────────────────────────────────────
|
||||
|
||||
async def _fetch_logprobs(
|
||||
session: aiohttp.ClientSession,
|
||||
teacher_url: str,
|
||||
full_ids: list[int],
|
||||
response_len: int,
|
||||
) -> list[float]:
|
||||
"""Call teacher sglang server and return per-token logprobs for the response portion."""
|
||||
payload = {
|
||||
"input_ids": full_ids,
|
||||
"sampling_params": {
|
||||
"temperature": 0,
|
||||
"max_new_tokens": 0,
|
||||
"skip_special_tokens": False,
|
||||
},
|
||||
"return_logprob": True,
|
||||
"logprob_start_len": 0,
|
||||
}
|
||||
async with session.post(teacher_url, json=payload) as resp:
|
||||
resp.raise_for_status()
|
||||
ret = await resp.json()
|
||||
|
||||
all_lps = ret["meta_info"]["input_token_logprobs"]
|
||||
response_lps = [float(item[0]) for item in all_lps[1:]][-response_len:]
|
||||
assert len(response_lps) == response_len, (
|
||||
f"Expected {response_len} logprobs, got {len(response_lps)}"
|
||||
)
|
||||
return response_lps
|
||||
|
||||
|
||||
async def _process_all(args, tokenizer, rows: list[dict]) -> list[list[float]]:
|
||||
"""Process all rows concurrently with a live progress bar, preserving order."""
|
||||
semaphore = asyncio.Semaphore(args.concurrency)
|
||||
connector = aiohttp.TCPConnector(limit=args.concurrency)
|
||||
results = [None] * len(rows)
|
||||
|
||||
async def bounded_fetch(idx: int, full_ids: list[int], response_len: int):
|
||||
async with semaphore:
|
||||
result = await _fetch_logprobs(session, args.teacher_url, full_ids, response_len)
|
||||
results[idx] = result
|
||||
pbar.update(1)
|
||||
|
||||
async with aiohttp.ClientSession(connector=connector) as session:
|
||||
with tqdm(total=len(rows), desc="[Phase 2] Teacher logprobs") as pbar:
|
||||
tasks = []
|
||||
for idx, row in enumerate(rows):
|
||||
meta = row["metadata"]
|
||||
prompt_ids = tokenizer.encode(row["prompt"], add_special_tokens=False)
|
||||
response_ids = [int(x) for x in meta["response_tokens"]]
|
||||
full_ids = prompt_ids + response_ids
|
||||
tasks.append(bounded_fetch(idx, full_ids, len(response_ids)))
|
||||
await asyncio.gather(*tasks)
|
||||
|
||||
return results
|
||||
|
||||
|
||||
def phase2_logprobs(args, intermediate_path: Path, output_path: Path):
|
||||
print(f"[Phase 2] Loading intermediate parquet: {intermediate_path}")
|
||||
df = pd.read_parquet(intermediate_path)
|
||||
rows = df.to_dict(orient="records")
|
||||
print(f"[Phase 2] Total rows: {len(rows)}")
|
||||
|
||||
print(f"[Phase 2] Loading tokenizer from {args.tokenizer_path}")
|
||||
tokenizer = AutoTokenizer.from_pretrained(args.tokenizer_path, trust_remote_code=True)
|
||||
|
||||
print(f"[Phase 2] Computing teacher logprobs via {args.teacher_url} "
|
||||
f"(concurrency={args.concurrency})")
|
||||
all_logprobs = asyncio.run(_process_all(args, tokenizer, rows))
|
||||
|
||||
for row, lps in zip(rows, all_logprobs):
|
||||
row["metadata"]["teacher_log_probs"] = lps
|
||||
|
||||
df_out = pd.DataFrame(rows)
|
||||
output_path.parent.mkdir(parents=True, exist_ok=True)
|
||||
df_out.to_parquet(output_path, index=False)
|
||||
print(f"[Phase 2] Saved to {output_path}")
|
||||
|
||||
# Sanity check
|
||||
df_check = pd.read_parquet(output_path)
|
||||
row0 = df_check.iloc[0]
|
||||
meta = row0["metadata"]
|
||||
print("\n[Phase 2] Sanity check row 0:")
|
||||
print(f" prompt[:80]: {row0['prompt'][:80]}")
|
||||
print(f" label: {row0['label']}")
|
||||
print(f" len(response_tokens): {len(meta['response_tokens'])}")
|
||||
print(f" len(teacher_log_probs): {len(meta['teacher_log_probs'])}")
|
||||
print(f" teacher_log_probs[:5]: {meta['teacher_log_probs'][:5]}")
|
||||
|
||||
|
||||
# ── Entry point ───────────────────────────────────────────────────────────────
|
||||
|
||||
def main():
|
||||
args = parse_args()
|
||||
|
||||
output_dir = Path(args.output_dir)
|
||||
input_stem = Path(args.input_parquet).stem
|
||||
intermediate_path = output_dir / f"{input_stem}-lightning-opd.parquet"
|
||||
output_path = output_dir / f"{input_stem}-lightning-opd-precomputed.parquet"
|
||||
|
||||
if args.compute_teacher_logprobs:
|
||||
if not intermediate_path.exists():
|
||||
print("[INFO] Intermediate parquet not found, running Phase 1 first.")
|
||||
phase1_tokenize(args, intermediate_path)
|
||||
phase2_logprobs(args, intermediate_path, output_path)
|
||||
else:
|
||||
phase1_tokenize(args, intermediate_path)
|
||||
print(f"\n[INFO] To add teacher logprobs, re-run with --compute-teacher-logprobs "
|
||||
f"after starting the teacher sglang server.")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
126
data_curation/run_curation.sh
Normal file
126
data_curation/run_curation.sh
Normal file
@@ -0,0 +1,126 @@
|
||||
#!/usr/bin/env bash
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
#
|
||||
# Launch data curation across multiple GPUs / nodes.
|
||||
#
|
||||
# Each GPU runs one independent vLLM worker that processes a disjoint shard
|
||||
# of the dataset. No torch.distributed communication is needed — each worker
|
||||
# is a standalone process with its own rank derived from environment variables.
|
||||
#
|
||||
# ── Single node, 8 GPUs (tp=1, 8 workers) ────────────────────────────────
|
||||
# bash data_curation/run_curation.sh \
|
||||
# --model Qwen/Qwen3-4B \
|
||||
# --input data.jsonl \
|
||||
# --output-dir output/ \
|
||||
# --num-gpus 8
|
||||
#
|
||||
# ── Single node, 2 GPUs (tp=2, 1 worker) ─────────────────────────────────
|
||||
# bash data_curation/run_curation.sh \
|
||||
# --model Qwen/Qwen3-8B \
|
||||
# --input data.jsonl \
|
||||
# --output-dir output/ \
|
||||
# --num-gpus 2 \
|
||||
# --tensor-parallel-size 2
|
||||
#
|
||||
# ── Multi-node (2 nodes × 8 GPUs, tp=1, 16 workers) ─────────────────────
|
||||
# # On node 0:
|
||||
# NODE_RANK=0 NUM_NODES=2 bash data_curation/run_curation.sh \
|
||||
# --model Qwen/Qwen3-4B \
|
||||
# --input data.jsonl \
|
||||
# --output-dir output/ \
|
||||
# --num-gpus 8
|
||||
#
|
||||
# # On node 1:
|
||||
# NODE_RANK=1 NUM_NODES=2 bash data_curation/run_curation.sh \
|
||||
# --model Qwen/Qwen3-4B \
|
||||
# --input data.jsonl \
|
||||
# --output-dir output/ \
|
||||
# --num-gpus 8
|
||||
#
|
||||
# Environment variables (optional):
|
||||
# NUM_NODES – total number of nodes (default: 1)
|
||||
# NODE_RANK – rank of this node (default: 0)
|
||||
# ──────────────────────────────────────────────────────────────────────────
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||||
|
||||
# ── Parse --num-gpus and --tensor-parallel-size from args ─────────────────
|
||||
NUM_GPUS=1
|
||||
TP=1
|
||||
PIPELINE_ARGS=()
|
||||
|
||||
while [[ $# -gt 0 ]]; do
|
||||
case "$1" in
|
||||
--num-gpus)
|
||||
NUM_GPUS="$2"; shift 2 ;;
|
||||
--tensor-parallel-size)
|
||||
TP="$2"; PIPELINE_ARGS+=("--tensor-parallel-size" "$2"); shift 2 ;;
|
||||
*)
|
||||
PIPELINE_ARGS+=("$1"); shift ;;
|
||||
esac
|
||||
done
|
||||
|
||||
# ── Compute worker layout ────────────────────────────────────────────────
|
||||
NUM_NODES="${NUM_NODES:-1}"
|
||||
NODE_RANK="${NODE_RANK:-0}"
|
||||
WORKERS_PER_NODE=$(( NUM_GPUS / TP ))
|
||||
WORLD_SIZE=$(( WORKERS_PER_NODE * NUM_NODES ))
|
||||
|
||||
echo "=== Data Curation Launch ==="
|
||||
echo " Nodes: ${NUM_NODES} (this node: ${NODE_RANK})"
|
||||
echo " GPUs per node: ${NUM_GPUS}"
|
||||
echo " TP size: ${TP}"
|
||||
echo " Workers per node: ${WORKERS_PER_NODE}"
|
||||
echo " World size: ${WORLD_SIZE}"
|
||||
echo " Pipeline args: ${PIPELINE_ARGS[*]}"
|
||||
echo "============================"
|
||||
|
||||
# ── Launch workers ───────────────────────────────────────────────────────
|
||||
PIDS=()
|
||||
for (( LOCAL=0; LOCAL<WORKERS_PER_NODE; LOCAL++ )); do
|
||||
GLOBAL_RANK=$(( NODE_RANK * WORKERS_PER_NODE + LOCAL ))
|
||||
GPU_START=$(( LOCAL * TP ))
|
||||
GPU_END=$(( GPU_START + TP - 1 ))
|
||||
|
||||
# Build CUDA_VISIBLE_DEVICES string, e.g. "0" or "2,3"
|
||||
GPUS=""
|
||||
for (( g=GPU_START; g<=GPU_END; g++ )); do
|
||||
[[ -n "$GPUS" ]] && GPUS="${GPUS},"
|
||||
GPUS="${GPUS}${g}"
|
||||
done
|
||||
|
||||
echo "[Node ${NODE_RANK}] Launching worker rank=${GLOBAL_RANK} on GPU(s) ${GPUS}"
|
||||
|
||||
CUDA_VISIBLE_DEVICES="${GPUS}" \
|
||||
RANK="${GLOBAL_RANK}" \
|
||||
WORLD_SIZE="${WORLD_SIZE}" \
|
||||
python "${SCRIPT_DIR}/pipeline.py" \
|
||||
--rank "${GLOBAL_RANK}" \
|
||||
--world-size "${WORLD_SIZE}" \
|
||||
"${PIPELINE_ARGS[@]}" \
|
||||
> >(sed "s/^/[rank${GLOBAL_RANK}] /") \
|
||||
2>&1 &
|
||||
|
||||
PIDS+=($!)
|
||||
done
|
||||
|
||||
# ── Wait for all workers ─────────────────────────────────────────────────
|
||||
echo "Waiting for ${#PIDS[@]} workers to finish..."
|
||||
FAILED=0
|
||||
for PID in "${PIDS[@]}"; do
|
||||
if ! wait "$PID"; then
|
||||
echo "Worker PID ${PID} failed!"
|
||||
FAILED=1
|
||||
fi
|
||||
done
|
||||
|
||||
if [[ $FAILED -eq 1 ]]; then
|
||||
echo "Some workers failed. Check logs above."
|
||||
exit 1
|
||||
fi
|
||||
|
||||
echo "All workers finished successfully."
|
||||
13
generation_config.json
Normal file
13
generation_config.json
Normal file
@@ -0,0 +1,13 @@
|
||||
{
|
||||
"bos_token_id": 151643,
|
||||
"do_sample": true,
|
||||
"eos_token_id": [
|
||||
151645,
|
||||
151643
|
||||
],
|
||||
"pad_token_id": 151643,
|
||||
"temperature": 0.6,
|
||||
"top_k": 20,
|
||||
"top_p": 0.95,
|
||||
"transformers_version": "4.52.4"
|
||||
}
|
||||
151388
merges.txt
Normal file
151388
merges.txt
Normal file
File diff suppressed because it is too large
Load Diff
3
model-00001-of-00002.safetensors
Normal file
3
model-00001-of-00002.safetensors
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:3d57b393ee9734c673d28da679609d593ade800fb4dfc7774a388da2a0be5654
|
||||
size 4967215360
|
||||
3
model-00002-of-00002.safetensors
Normal file
3
model-00002-of-00002.safetensors
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:2fd291e26278a84c71130076fd0ce28ed048b625e6725e893fdaf4a9dc6c40ee
|
||||
size 3855679144
|
||||
406
model.safetensors.index.json
Normal file
406
model.safetensors.index.json
Normal file
@@ -0,0 +1,406 @@
|
||||
{
|
||||
"metadata": {
|
||||
"total_size": 8822848512
|
||||
},
|
||||
"weight_map": {
|
||||
"lm_head.weight": "model-00002-of-00002.safetensors",
|
||||
"model.embed_tokens.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.0.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.1.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.10.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.11.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.12.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.13.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.14.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.15.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.16.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.17.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.18.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.19.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.2.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.20.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.20.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.20.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.21.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.21.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.22.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.23.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.24.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.25.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.26.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.27.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.28.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.29.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.3.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.3.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.30.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.30.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.31.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.32.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.33.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.34.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.input_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.mlp.down_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.mlp.gate_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.mlp.up_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.post_attention_layernorm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.k_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.k_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.o_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.q_norm.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.q_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.35.self_attn.v_proj.weight": "model-00002-of-00002.safetensors",
|
||||
"model.layers.4.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.4.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.5.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.6.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.7.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.8.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.input_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.down_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.gate_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.mlp.up_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.post_attention_layernorm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.k_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.k_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.o_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.q_norm.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.q_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.layers.9.self_attn.v_proj.weight": "model-00001-of-00002.safetensors",
|
||||
"model.norm.weight": "model-00002-of-00002.safetensors"
|
||||
}
|
||||
}
|
||||
59
pyproject.toml
Normal file
59
pyproject.toml
Normal file
@@ -0,0 +1,59 @@
|
||||
[build-system]
|
||||
requires = [
|
||||
"packaging",
|
||||
"setuptools >= 49.4.0",
|
||||
"wheel",
|
||||
]
|
||||
build-backend = "setuptools.build_meta"
|
||||
|
||||
[tool.isort]
|
||||
profile = "black" # black-compatible
|
||||
line_length = 119 # should match black parameters
|
||||
ignore_whitespace = true # ignore whitespace for compatibility with the initial style
|
||||
py_version = 310 # python 3.10 as a target version
|
||||
sections = ["FUTURE", "STDLIB", "THIRDPARTY", "FIRSTPARTY", "LOCALFOLDER"]
|
||||
default_section = "THIRDPARTY"
|
||||
extend_skip = ["setup.py", "docs/source/conf.py"]
|
||||
known_first_party = ["slime", "slime_plugins"]
|
||||
known_third_party = ["megatron", "wandb", "ray", "transformers"]
|
||||
src_paths = ["slime", "slime_plugins"]
|
||||
|
||||
|
||||
[tool.black]
|
||||
line_length = 119
|
||||
|
||||
[tool.ruff]
|
||||
line-length = 119
|
||||
|
||||
[tool.pytest.ini_options]
|
||||
# durations=0 will display all tests execution time, sorted in ascending order starting from from the slowest one.
|
||||
# -vv will also display tests with duration = 0.00s
|
||||
addopts = "--verbose --pyargs --durations=0 --strict-markers" # always add these arguments to pytest
|
||||
testpaths = ["./tests"] # must be an explicit path to avoid importing another "tests" module
|
||||
# directories to ignore when discovering tests
|
||||
norecursedirs = [
|
||||
"external",
|
||||
"examples",
|
||||
"docs",
|
||||
"scripts",
|
||||
"tools",
|
||||
"tutorials",
|
||||
"*.egg",
|
||||
".*",
|
||||
"_darcs",
|
||||
"build",
|
||||
"CVS",
|
||||
"dist",
|
||||
"venv",
|
||||
"{arch}",
|
||||
]
|
||||
# markers to select tests, use `pytest --markers` to see all available markers, `pytest -m "<marker>"` to select tests
|
||||
markers = [
|
||||
"unit: marks unit test, i.e. testing a single, well isolated functionality (deselect with '-m \"not unit\"')",
|
||||
"integration: marks test checking the elements when integrated into subsystems (deselect with '-m \"not integration\"')",
|
||||
"system: marks test working at the highest integration level (deselect with '-m \"not system\"')",
|
||||
"acceptance: marks test checking whether the developed product/model passes the user defined acceptance criteria (deselect with '-m \"not acceptance\"')",
|
||||
"docs: mark tests related to documentation (deselect with '-m \"not docs\"')",
|
||||
"skipduringci: marks tests that are skipped ci as they are addressed by Jenkins jobs but should be run to test user setups",
|
||||
"pleasefixme: marks tests that are broken and need fixing",
|
||||
]
|
||||
19
requirements.txt
Normal file
19
requirements.txt
Normal file
@@ -0,0 +1,19 @@
|
||||
accelerate
|
||||
blobfile
|
||||
datasets
|
||||
httpx[http2]
|
||||
mcp[cli]
|
||||
megatron-bridge @ git+https://github.com/fzyzcjy/Megatron-Bridge.git@dev_rl
|
||||
memray # needed for debugging (but is lightweight), we can put it to dev mode when using pyproject.toml
|
||||
nvidia-modelopt[torch]>=0.37.0 # needed by megatron-bridge
|
||||
omegaconf
|
||||
pillow
|
||||
pylatexenc
|
||||
pyyaml
|
||||
ray[default]
|
||||
ring_flash_attn
|
||||
sglang-router>=0.2.3
|
||||
tensorboard
|
||||
transformers
|
||||
wandb
|
||||
liger_kernel
|
||||
10
run_docker.sh
Normal file
10
run_docker.sh
Normal file
@@ -0,0 +1,10 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
docker run -it --gpus all \
|
||||
--shm-size=64g \
|
||||
-v $(pwd):/workspace/Lightning-OPD \
|
||||
-v $HOME/.cache:$HOME/.cache \
|
||||
-w /workspace/Lightning-OPD \
|
||||
tonyhao96/jetmoe:v0.2 \
|
||||
bash
|
||||
42
scripts/collect_rollouts.sh
Normal file
42
scripts/collect_rollouts.sh
Normal file
@@ -0,0 +1,42 @@
|
||||
#!/usr/bin/env bash
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
# Step 3: Collect student rollouts on OPD prompts.
|
||||
#
|
||||
# Uses data_curation/ to run the SFT model on OPD prompts (e.g. DAPO-Math-17k)
|
||||
# and collect response rollouts for Lightning OPD data preparation.
|
||||
#
|
||||
# Required environment variables:
|
||||
# SFT_CHECKPOINT - Path to the SFT model checkpoint
|
||||
# OPD_PROMPTS - Path to the OPD prompt dataset (.jsonl or .parquet)
|
||||
# OUTPUT_DIR - Directory for collected rollout data
|
||||
#
|
||||
# Optional:
|
||||
# NUM_GPUS - Number of GPUs to use (default: 8)
|
||||
# TP_SIZE - Tensor parallel size per worker (default: 1)
|
||||
#
|
||||
# Extra args are passed through to data_curation/pipeline.py, e.g.:
|
||||
# bash scripts/collect_rollouts.sh --num-samples 10
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
: "${SFT_CHECKPOINT:?Set SFT_CHECKPOINT to the SFT model path}"
|
||||
: "${OPD_PROMPTS:?Set OPD_PROMPTS to the OPD prompt dataset path}"
|
||||
: "${OUTPUT_DIR:?Set OUTPUT_DIR for collected rollout data}"
|
||||
|
||||
# Resolve to absolute paths (workers may run from different cwd)
|
||||
SFT_CHECKPOINT="$(cd "$(dirname "${SFT_CHECKPOINT}")" && pwd)/$(basename "${SFT_CHECKPOINT}")"
|
||||
OPD_PROMPTS="$(cd "$(dirname "${OPD_PROMPTS}")" && pwd)/$(basename "${OPD_PROMPTS}")"
|
||||
OUTPUT_DIR="$(mkdir -p "${OUTPUT_DIR}" && cd "${OUTPUT_DIR}" && pwd)"
|
||||
|
||||
NUM_GPUS="${NUM_GPUS:-8}"
|
||||
TP_SIZE="${TP_SIZE:-1}"
|
||||
|
||||
bash data_curation/run_curation.sh \
|
||||
--model "${SFT_CHECKPOINT}" \
|
||||
--input "${OPD_PROMPTS}" \
|
||||
--output-dir "${OUTPUT_DIR}" \
|
||||
--num-gpus "${NUM_GPUS}" \
|
||||
--tensor-parallel-size "${TP_SIZE}" \
|
||||
"$@"
|
||||
27
scripts/convert_megatron_to_hf.sh
Normal file
27
scripts/convert_megatron_to_hf.sh
Normal file
@@ -0,0 +1,27 @@
|
||||
#!/bin/bash
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
#
|
||||
# Convert a Megatron torch_dist checkpoint to HuggingFace format.
|
||||
#
|
||||
# Required env vars:
|
||||
# MEGATRON_CKPT_DIR - path to the Megatron checkpoint directory (e.g., /root/models/<name>_ckpt__<config>/iter_0000150)
|
||||
# HF_OUTPUT_DIR - path to save the converted HuggingFace model
|
||||
# ORIGIN_HF_DIR - path to the original HuggingFace model (for config.json, tokenizer, etc.)
|
||||
#
|
||||
# Example:
|
||||
# MEGATRON_CKPT_DIR=/root/models/Qwen3-4B-Base-sft_ckpt__qwen3-4b-lightning-opd/iter_0000150 \
|
||||
# HF_OUTPUT_DIR=checkpoints/qwen3-4b-lightning-opd-hf \
|
||||
# ORIGIN_HF_DIR=checkpoints/qwen3-4b-base-sft-qwen3-8b/<your-sft-checkpoint> \
|
||||
# bash scripts/convert_megatron_to_hf.sh
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
: "${MEGATRON_CKPT_DIR:?Please set MEGATRON_CKPT_DIR}"
|
||||
: "${HF_OUTPUT_DIR:?Please set HF_OUTPUT_DIR}"
|
||||
: "${ORIGIN_HF_DIR:?Please set ORIGIN_HF_DIR}"
|
||||
|
||||
python tools/convert_torch_dist_to_hf.py \
|
||||
--input-dir "${MEGATRON_CKPT_DIR}" \
|
||||
--output-dir "${HF_OUTPUT_DIR}" \
|
||||
--origin-hf-dir "${ORIGIN_HF_DIR}"
|
||||
53
scripts/eval_aime2024.sh
Normal file
53
scripts/eval_aime2024.sh
Normal file
@@ -0,0 +1,53 @@
|
||||
export MKL_THREADING_LAYER=GNU
|
||||
export MKL_SERVICE_FORCE_INTEL=0
|
||||
export OMP_NUM_THREADS=1
|
||||
|
||||
# CUDA_VISIBLE_DEVICES=0 python tools/eval_aime2024_vllm.py \
|
||||
# --model /mnt/disk1/yihao/Lightning-OPD/model_weights/qwen3-8b \
|
||||
# --num-gpus 1 \
|
||||
# --n-samples 1 \
|
||||
# --temperature 0.0 \
|
||||
# --top-p 1.0 \
|
||||
# --max-tokens 32768 \
|
||||
# --prompt-template paper \
|
||||
# --hf-cache /mnt/disk1/yihao/hf_cache \
|
||||
# --output outputs/aime2024_qwen3_8b_paper_n1_32k.jsonl
|
||||
|
||||
# CUDA_VISIBLE_DEVICES=0 python tools/eval_aime2024_vllm.py \
|
||||
# --model /mnt/disk1/yihao/Lightning-OPD/checkpoints/qwen3-4b-poe-distill-lora-opd-ppo-clip-locking-b-5-self-distill-100 \
|
||||
# --num-gpus 1 \
|
||||
# --n-samples 1 \
|
||||
# --temperature 0.0 \
|
||||
# --top-p 1.0 \
|
||||
# --max-tokens 32768 \
|
||||
# --prompt-template paper \
|
||||
# --hf-cache /mnt/disk1/yihao/hf_cache \
|
||||
# --output outputs/aime2024_qwen3_8b_paper_n1_32k.jsonl
|
||||
|
||||
|
||||
CUDA_VISIBLE_DEVICES=4,5,6,7 python tools/eval_aime2024_vllm.py \
|
||||
--model /mnt/disk1/yihao/Lightning-OPD/checkpoints/qwen3-4b-poe-distill-lora-opd-ppo-clip-locking-b-5-self-distill-100 \
|
||||
--num-gpus 4 \
|
||||
--n-samples 32 \
|
||||
--temperature 0.6 \
|
||||
--top-p 0.95 \
|
||||
--max-tokens 32768 \
|
||||
--prompt-template paper \
|
||||
--hf-cache /mnt/disk1/yihao/hf_cache \
|
||||
--output outputs/aime2024_qwen3_4b_poe_distill_lora_paper_n1_32k.jsonl \
|
||||
--enable-thinking
|
||||
|
||||
|
||||
|
||||
# CUDA_VISIBLE_DEVICES=4,5,6,7 python tools/eval_aime2024_vllm.py \
|
||||
# --model /mnt/disk1/yihao/Lightning-OPD/checkpoints/qwen3-4b-poe-distill-lora-opd-ppo-clip-60 \
|
||||
# --num-gpus 4 \
|
||||
# --n-samples 32 \
|
||||
# --temperature 0.6 \
|
||||
# --top-p 0.95 \
|
||||
# --max-tokens 32768 \
|
||||
# --enable-thinking \
|
||||
# --prompt-template paper \
|
||||
# --hf-cache /mnt/disk1/yihao/hf_cache \
|
||||
# --output outputs/aime2024_qwen3_4b_lightning_opd_paper_n32_32k.jsonl \
|
||||
|
||||
23
scripts/eval_aime2025.sh
Normal file
23
scripts/eval_aime2025.sh
Normal file
@@ -0,0 +1,23 @@
|
||||
CUDA_VISIBLE_DEVICES=4,5,6,7 python tools/eval_aime2025_vllm.py \
|
||||
--model /mnt/disk1/yihao/Lightning-OPD/checkpoints/qwen3-4b-lightning-opd-hf \
|
||||
--num-gpus 4 \
|
||||
--prompt-template paper \
|
||||
--n-samples 32 \
|
||||
--temperature 0.6 \
|
||||
--top-p 0.95 \
|
||||
--max-tokens 32768 \
|
||||
--enable-thinking \
|
||||
--output outputs/aime2025_qwen3_4b_lightning_opd.jsonl
|
||||
|
||||
|
||||
# CUDA_VISIBLE_DEVICES=0,2 python tools/eval_aime2025_vllm.py \
|
||||
# --model /mnt/disk1/yihao/Lightning-OPD/checkpoints/qwen3-4b-lightning-opd-hf \
|
||||
# --num-gpus 2 \
|
||||
# --n-samples 1 \
|
||||
# --temperature 0.0 \
|
||||
# --top-p 1.0 \
|
||||
# --max-tokens 32768 \
|
||||
# --prompt-template paper \
|
||||
# --hf-cache /mnt/disk1/yihao/hf_cache \
|
||||
# --output outputs/aime2024_qwen3_4b_poe_distill_lora_paper_n1_32k.jsonl \
|
||||
# --enable-thinking
|
||||
10
scripts/eval_hmmt25.sh
Normal file
10
scripts/eval_hmmt25.sh
Normal file
@@ -0,0 +1,10 @@
|
||||
CUDA_VISIBLE_DEVICES=4,5,6,7 python tools/eval_hmmt2025_vllm.py \
|
||||
--model /mnt/disk1/yihao/Lightning-OPD/checkpoints/qwen3-4b-lightning-opd-hf \
|
||||
--num-gpus 4 \
|
||||
--prompt-template paper \
|
||||
--n-samples 32 \
|
||||
--temperature 0.6 \
|
||||
--top-p 0.95 \
|
||||
--max-tokens 32768 \
|
||||
--enable-thinking \
|
||||
--output outputs/hmmt_feb_2025_qwen3_4b_lightning_opd.jsonl
|
||||
59
scripts/generate_sft_data.sh
Normal file
59
scripts/generate_sft_data.sh
Normal file
@@ -0,0 +1,59 @@
|
||||
#!/usr/bin/env bash
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
# Step 1: Generate SFT data using the teacher model.
|
||||
#
|
||||
# Uses data_curation/ to run the teacher model on OpenThoughts-3 prompts
|
||||
# and generate response trajectories for SFT training.
|
||||
#
|
||||
# Required environment variables:
|
||||
# TEACHER_MODEL - HuggingFace model name or path (e.g. Qwen/Qwen3-8B)
|
||||
# SFT_PROMPTS - Path to the prompt dataset (.jsonl or .parquet)
|
||||
# OUTPUT_DIR - Directory for generated SFT data
|
||||
#
|
||||
# Optional:
|
||||
# NUM_GPUS - Number of GPUs to use (default: 8)
|
||||
# TP_SIZE - Tensor parallel size per worker (default: 1)
|
||||
#
|
||||
# Extra args are passed through to data_curation/pipeline.py, e.g.:
|
||||
# bash scripts/generate_sft_data.sh --num-samples 10
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
# ---- CUDA / FlashInfer build environment ----
|
||||
if [ -n "${CONDA_PREFIX:-}" ]; then
|
||||
export CUDA_HOME="${CUDA_HOME:-$CONDA_PREFIX}"
|
||||
export CUDA_PATH="${CUDA_PATH:-$CONDA_PREFIX}"
|
||||
export CUDACXX="${CUDACXX:-$CONDA_PREFIX/bin/nvcc}"
|
||||
|
||||
export PATH="$CONDA_PREFIX/bin:$PATH"
|
||||
export LD_LIBRARY_PATH="$CONDA_PREFIX/lib:$CONDA_PREFIX/lib64:/usr/lib/x86_64-linux-gnu:${LD_LIBRARY_PATH:-}"
|
||||
export LIBRARY_PATH="/usr/lib/x86_64-linux-gnu:${LIBRARY_PATH:-}"
|
||||
fi
|
||||
|
||||
echo "Using nvcc: $(which nvcc)"
|
||||
nvcc --version || true
|
||||
echo "CUDA_HOME=${CUDA_HOME:-}"
|
||||
echo "CUDACXX=${CUDACXX:-}"
|
||||
echo "LIBRARY_PATH=${LIBRARY_PATH:-}"
|
||||
# ---------------------------------------------
|
||||
|
||||
: "${TEACHER_MODEL:?Set TEACHER_MODEL (e.g. Qwen/Qwen3-8B)}"
|
||||
: "${SFT_PROMPTS:?Set SFT_PROMPTS to the prompt dataset path}"
|
||||
: "${OUTPUT_DIR:?Set OUTPUT_DIR for generated SFT data}"
|
||||
|
||||
# Resolve to absolute paths (workers may run from different cwd)
|
||||
SFT_PROMPTS="$(cd "$(dirname "${SFT_PROMPTS}")" && pwd)/$(basename "${SFT_PROMPTS}")"
|
||||
OUTPUT_DIR="$(mkdir -p "${OUTPUT_DIR}" && cd "${OUTPUT_DIR}" && pwd)"
|
||||
|
||||
NUM_GPUS="${NUM_GPUS:-8}"
|
||||
TP_SIZE="${TP_SIZE:-1}"
|
||||
|
||||
bash data_curation/run_curation.sh \
|
||||
--model "${TEACHER_MODEL}" \
|
||||
--input "${SFT_PROMPTS}" \
|
||||
--output-dir "${OUTPUT_DIR}" \
|
||||
--num-gpus "${NUM_GPUS}" \
|
||||
--tensor-parallel-size "${TP_SIZE}" \
|
||||
"$@"
|
||||
5
scripts/merge_poe_lora.sh
Normal file
5
scripts/merge_poe_lora.sh
Normal file
@@ -0,0 +1,5 @@
|
||||
python tools/merge_poe_lora.py \
|
||||
--base-model checkpoints/qwen3-4b-base-sft-qwen3-8b \
|
||||
--adapter checkpoints/qwen3-4b-poe-distill-lora-opd-ppo-clip-locking-b-5-self-distill/checkpoint-100 \
|
||||
--output-dir checkpoints/qwen3-4b-poe-distill-lora-opd-ppo-clip-locking-b-5-self-distill-100 \
|
||||
--dtype bfloat16
|
||||
29
scripts/precompute_teacher_logprobs_4b.sh
Normal file
29
scripts/precompute_teacher_logprobs_4b.sh
Normal file
@@ -0,0 +1,29 @@
|
||||
#!/usr/bin/env bash
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
# Precompute teacher logprobs for Lightning OPD (4B scale, teacher=Qwen3-8B).
|
||||
#
|
||||
# Required environment variables:
|
||||
# SFT_CHECKPOINT - Path to the SFT checkpoint (used as tokenizer)
|
||||
# ROLLOUT_PARQUET - Path to the student rollout parquet file
|
||||
# OUTPUT_DIR - Directory for the output parquet with teacher logprobs
|
||||
#
|
||||
# This script starts a Qwen3-8B teacher server, then runs Phase 1+2 of
|
||||
# prepare_lightning_opd.py to tokenize and precompute teacher logprobs.
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
: "${SFT_CHECKPOINT:?Set SFT_CHECKPOINT to the SFT model path}"
|
||||
: "${ROLLOUT_PARQUET:?Set ROLLOUT_PARQUET to the student rollout parquet}"
|
||||
: "${OUTPUT_DIR:?Set OUTPUT_DIR for the output parquet}"
|
||||
|
||||
# Start teacher server
|
||||
bash scripts/serve_teacher_8b.sh
|
||||
|
||||
python3 data_curation/prepare_lightning_opd.py \
|
||||
--tokenizer-path "${SFT_CHECKPOINT}" \
|
||||
--input-parquet "${ROLLOUT_PARQUET}" \
|
||||
--output-dir "${OUTPUT_DIR}" \
|
||||
--compute-teacher-logprobs \
|
||||
--teacher-url http://127.0.0.1:13141/generate
|
||||
29
scripts/precompute_teacher_logprobs_8b.sh
Normal file
29
scripts/precompute_teacher_logprobs_8b.sh
Normal file
@@ -0,0 +1,29 @@
|
||||
#!/usr/bin/env bash
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
# Precompute teacher logprobs for Lightning OPD (8B scale, teacher=Qwen3-32B).
|
||||
#
|
||||
# Required environment variables:
|
||||
# SFT_CHECKPOINT - Path to the SFT checkpoint (used as tokenizer)
|
||||
# ROLLOUT_PARQUET - Path to the student rollout parquet file
|
||||
# OUTPUT_DIR - Directory for the output parquet with teacher logprobs
|
||||
#
|
||||
# This script starts a Qwen3-32B teacher server, then runs Phase 1+2 of
|
||||
# prepare_lightning_opd.py to tokenize and precompute teacher logprobs.
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
: "${SFT_CHECKPOINT:?Set SFT_CHECKPOINT to the SFT model path}"
|
||||
: "${ROLLOUT_PARQUET:?Set ROLLOUT_PARQUET to the student rollout parquet}"
|
||||
: "${OUTPUT_DIR:?Set OUTPUT_DIR for the output parquet}"
|
||||
|
||||
# Start teacher server
|
||||
bash scripts/serve_teacher_32b.sh
|
||||
|
||||
python3 data_curation/prepare_lightning_opd.py \
|
||||
--tokenizer-path "${SFT_CHECKPOINT}" \
|
||||
--input-parquet "${ROLLOUT_PARQUET}" \
|
||||
--output-dir "${OUTPUT_DIR}" \
|
||||
--compute-teacher-logprobs \
|
||||
--teacher-url http://127.0.0.1:13141/generate
|
||||
150
scripts/prepare_sft_prompts.py
Normal file
150
scripts/prepare_sft_prompts.py
Normal file
@@ -0,0 +1,150 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
"""
|
||||
Convert HuggingFace OpenThoughts3-1.2M dataset to a prompt-only JSONL file
|
||||
for SFT data generation (Step 1).
|
||||
|
||||
Extracts the prompt (user messages) from each sample and writes to JSONL.
|
||||
Optionally samples a subset (default 300K) to reduce compute cost.
|
||||
|
||||
Usage:
|
||||
python scripts/prepare_sft_prompts.py \
|
||||
--output data/prompts/openthoughts3_300k.jsonl \
|
||||
--num-samples 300000
|
||||
|
||||
# Use a local parquet file instead of downloading from HF
|
||||
python scripts/prepare_sft_prompts.py \
|
||||
--input data/prompts/local.parquet \
|
||||
--output data/prompts/openthoughts3_300k.jsonl
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import random
|
||||
|
||||
|
||||
def parse_args():
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Extract prompts from OpenThoughts3-1.2M for SFT data generation."
|
||||
)
|
||||
parser.add_argument(
|
||||
"--input", type=str, default=None,
|
||||
help="Path to a local parquet/jsonl file. If not set, downloads from HuggingFace.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--hf-dataset", type=str, default="open-thoughts/OpenThoughts3-1.2M",
|
||||
help="HuggingFace dataset name (default: open-thoughts/OpenThoughts3-1.2M).",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output", type=str, required=True,
|
||||
help="Output JSONL file path.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--num-samples", type=int, default=300000,
|
||||
help="Number of samples to keep (default: 300000). Set to 0 for all.",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--seed", type=int, default=42,
|
||||
help="Random seed for sampling (default: 42).",
|
||||
)
|
||||
return parser.parse_args()
|
||||
|
||||
|
||||
def extract_prompt(sample):
|
||||
"""Extract the prompt (non-assistant messages) from a sample.
|
||||
|
||||
Supports two common formats:
|
||||
1. {"conversations": [{"from": "human", "value": ...}, ...]} (sharegpt)
|
||||
2. {"prompt": [{"role": "user", "content": ...}, ...]} (chat messages)
|
||||
"""
|
||||
if "conversations" in sample:
|
||||
messages = []
|
||||
for turn in sample["conversations"]:
|
||||
role = turn.get("from", turn.get("role", ""))
|
||||
content = turn.get("value", turn.get("content", ""))
|
||||
if role in ("human", "user"):
|
||||
messages.append({"role": "user", "content": content})
|
||||
elif role == "system":
|
||||
messages.append({"role": "system", "content": content})
|
||||
if messages:
|
||||
return {"prompt": messages}
|
||||
|
||||
if "prompt" in sample:
|
||||
if isinstance(sample["prompt"], list):
|
||||
return {"prompt": sample["prompt"]}
|
||||
elif isinstance(sample["prompt"], str):
|
||||
return {"prompt": [{"role": "user", "content": sample["prompt"]}]}
|
||||
|
||||
if "messages" in sample:
|
||||
messages = [
|
||||
{"role": m["role"], "content": m["content"]}
|
||||
for m in sample["messages"]
|
||||
if m["role"] != "assistant"
|
||||
]
|
||||
if messages:
|
||||
return {"prompt": messages}
|
||||
|
||||
return None
|
||||
|
||||
|
||||
def load_dataset_from_hf(dataset_name):
|
||||
"""Load dataset from HuggingFace."""
|
||||
from datasets import load_dataset
|
||||
print(f"Loading dataset from HuggingFace: {dataset_name}")
|
||||
ds = load_dataset(dataset_name, split="train")
|
||||
return ds
|
||||
|
||||
|
||||
def load_dataset_from_file(path):
|
||||
"""Load dataset from local file (parquet or jsonl)."""
|
||||
import pandas as pd
|
||||
print(f"Loading dataset from local file: {path}")
|
||||
if path.endswith(".parquet"):
|
||||
df = pd.read_parquet(path)
|
||||
return df.to_dict("records")
|
||||
elif path.endswith(".jsonl"):
|
||||
with open(path) as f:
|
||||
return [json.loads(line) for line in f]
|
||||
else:
|
||||
raise ValueError(f"Unsupported format: {path}")
|
||||
|
||||
|
||||
def main():
|
||||
args = parse_args()
|
||||
random.seed(args.seed)
|
||||
|
||||
# Load dataset
|
||||
if args.input:
|
||||
samples = load_dataset_from_file(args.input)
|
||||
else:
|
||||
samples = load_dataset_from_hf(args.hf_dataset)
|
||||
|
||||
print(f"Total samples: {len(samples)}")
|
||||
|
||||
# Sample subset
|
||||
if args.num_samples > 0 and args.num_samples < len(samples):
|
||||
indices = random.sample(range(len(samples)), args.num_samples)
|
||||
indices.sort()
|
||||
samples = [samples[i] for i in indices]
|
||||
print(f"Sampled {args.num_samples} samples")
|
||||
|
||||
# Extract prompts
|
||||
from tqdm import tqdm
|
||||
written = 0
|
||||
skipped = 0
|
||||
with open(args.output, "w") as f:
|
||||
for sample in tqdm(samples, desc="Extracting prompts"):
|
||||
prompt_item = extract_prompt(sample)
|
||||
if prompt_item and len(prompt_item["prompt"]) > 0:
|
||||
f.write(json.dumps(prompt_item) + "\n")
|
||||
written += 1
|
||||
else:
|
||||
skipped += 1
|
||||
|
||||
print(f"Written: {written}, Skipped: {skipped}")
|
||||
print(f"Output: {args.output}")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
126
scripts/run_poe_distill_qwen3_4b_lora.sh
Normal file
126
scripts/run_poe_distill_qwen3_4b_lora.sh
Normal file
@@ -0,0 +1,126 @@
|
||||
#!/usr/bin/env bash
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
# Starter launcher for PoE / sampled-token OPD LoRA training with:
|
||||
# 1) hold-then-transition beta schedule
|
||||
# 2) optional hold-then-transition learning-rate schedule
|
||||
#
|
||||
# Default schedule below:
|
||||
# steps 0-99: beta = 1.0, lr = 2e-6
|
||||
# steps 100-110: beta 1.0 -> 0.5, lr 2e-6 -> 1e-7
|
||||
# steps 111+: beta = 0.5, lr = 1e-7
|
||||
#
|
||||
# Note: these are optimizer global steps, not micro-batch steps.
|
||||
|
||||
set -euo pipefail
|
||||
|
||||
: "${SFT_CHECKPOINT:?Set SFT_CHECKPOINT to the Qwen3-4B SFT checkpoint}"
|
||||
|
||||
# Path to the patched Python trainer. If you copy the Python file into tools/,
|
||||
# leave this default; otherwise pass SCRIPT_PATH=/path/to/file.py.
|
||||
SCRIPT_PATH="${SCRIPT_PATH:-tools/train_poe_distill_lora.py}"
|
||||
|
||||
TEACHER_MODEL="${TEACHER_MODEL:-model_weights/qwen3-8b}"
|
||||
TRAIN_DATA="${TRAIN_DATA:-data/rollouts/dapo-math-17k-qwen3-4b-sft-rollouts.parquet}"
|
||||
OUTPUT_DIR="${OUTPUT_DIR:-checkpoints/qwen3-4b-poe-distill-lora-opd-hold-beta-lr}"
|
||||
|
||||
CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES:-0}"
|
||||
NPROC_PER_NODE="${NPROC_PER_NODE:-1}"
|
||||
|
||||
# Training length.
|
||||
MAX_STEPS="${MAX_STEPS:-120}"
|
||||
SAVE_STEPS="${SAVE_STEPS:-10}"
|
||||
LOGGING_STEPS="${LOGGING_STEPS:-1}"
|
||||
|
||||
# Beta schedule: hold beta_start, transition to beta_end.
|
||||
BETA_START="${BETA_START:-1.0}"
|
||||
BETA_END="${BETA_END:-0.5}"
|
||||
BETA_HOLD_STEPS="${BETA_HOLD_STEPS:-100}"
|
||||
BETA_TRANSITION_STEPS="${BETA_TRANSITION_STEPS:-10}"
|
||||
HOLD_TRANSITION_SCHEDULE="${HOLD_TRANSITION_SCHEDULE:-linear}" # linear or cosine
|
||||
|
||||
# LR schedule: hold lr_start, transition to lr_end.
|
||||
# When LR_END is non-empty, use constant HF scheduler and let the callback set LR.
|
||||
LEARNING_RATE="${LEARNING_RATE:-2e-6}"
|
||||
LR_START="${LR_START:-${LEARNING_RATE}}"
|
||||
LR_END="${LR_END:-1e-7}"
|
||||
LR_HOLD_STEPS="${LR_HOLD_STEPS:-${BETA_HOLD_STEPS}}"
|
||||
LR_TRANSITION_STEPS="${LR_TRANSITION_STEPS:-${BETA_TRANSITION_STEPS}}"
|
||||
|
||||
# Loss / OPD behavior.
|
||||
LOSS_TYPE="${LOSS_TYPE:-sampled_token}"
|
||||
ADVANTAGE_NORMALIZATION="${ADVANTAGE_NORMALIZATION:-none}"
|
||||
ADVANTAGE_CLIP="${ADVANTAGE_CLIP:-10.0}"
|
||||
USE_PPO_CLIP="${USE_PPO_CLIP:-1}"
|
||||
PPO_CLIP_LOW="${PPO_CLIP_LOW:-0.2}"
|
||||
PPO_CLIP_HIGH="${PPO_CLIP_HIGH:-0.2}"
|
||||
SAMPLED_LOSS_REDUCTION="${SAMPLED_LOSS_REDUCTION:-per_sample}"
|
||||
POSITIVE_ADVANTAGES_ONLY="${POSITIVE_ADVANTAGES_ONLY:-0}"
|
||||
|
||||
# Model / optimizer defaults.
|
||||
MAX_LENGTH="${MAX_LENGTH:-4096}"
|
||||
DISTILL_CHUNK_SIZE="${DISTILL_CHUNK_SIZE:-128}"
|
||||
PER_DEVICE_TRAIN_BATCH_SIZE="${PER_DEVICE_TRAIN_BATCH_SIZE:-2}"
|
||||
GRADIENT_ACCUMULATION_STEPS="${GRADIENT_ACCUMULATION_STEPS:-8}"
|
||||
WEIGHT_DECAY="${WEIGHT_DECAY:-0.1}"
|
||||
ADAM_BETA1="${ADAM_BETA1:-0.9}"
|
||||
ADAM_BETA2="${ADAM_BETA2:-0.98}"
|
||||
WARMUP_RATIO="${WARMUP_RATIO:-0.0}"
|
||||
LR_SCHEDULER_TYPE="${LR_SCHEDULER_TYPE:-constant}"
|
||||
FREEZE_LORA_B_AFTER_STEP="${FREEZE_LORA_B_AFTER_STEP:-99999}"
|
||||
|
||||
PPO_CLIP_ARGS=()
|
||||
if [[ "${USE_PPO_CLIP}" == "1" || "${USE_PPO_CLIP}" == "true" || "${USE_PPO_CLIP}" == "True" ]]; then
|
||||
PPO_CLIP_ARGS+=(--use-ppo-clip)
|
||||
fi
|
||||
|
||||
POS_ADV_ARGS=()
|
||||
if [[ "${POSITIVE_ADVANTAGES_ONLY}" == "1" || "${POSITIVE_ADVANTAGES_ONLY}" == "true" || "${POSITIVE_ADVANTAGES_ONLY}" == "True" ]]; then
|
||||
POS_ADV_ARGS+=(--positive-advantages-only)
|
||||
fi
|
||||
|
||||
LR_ARGS=(--lr-start "${LR_START}")
|
||||
if [[ -n "${LR_END}" ]]; then
|
||||
LR_ARGS+=(
|
||||
--lr-end "${LR_END}"
|
||||
--lr-hold-steps "${LR_HOLD_STEPS}"
|
||||
--lr-transition-steps "${LR_TRANSITION_STEPS}"
|
||||
)
|
||||
fi
|
||||
|
||||
CUDA_VISIBLE_DEVICES="${CUDA_VISIBLE_DEVICES}" torchrun --standalone --nproc_per_node="${NPROC_PER_NODE}" "${SCRIPT_PATH}" \
|
||||
--student-model "${SFT_CHECKPOINT}" \
|
||||
--teacher-model "${TEACHER_MODEL}" \
|
||||
--train-data "${TRAIN_DATA}" \
|
||||
--output-dir "${OUTPUT_DIR}" \
|
||||
--alpha "${ALPHA:-1.0}" \
|
||||
--beta-start "${BETA_START}" \
|
||||
--beta-end "${BETA_END}" \
|
||||
--beta-hold-steps "${BETA_HOLD_STEPS}" \
|
||||
--beta-transition-steps "${BETA_TRANSITION_STEPS}" \
|
||||
--hold-transition-schedule "${HOLD_TRANSITION_SCHEDULE}" \
|
||||
"${LR_ARGS[@]}" \
|
||||
--loss-type "${LOSS_TYPE}" \
|
||||
--advantage-normalization "${ADVANTAGE_NORMALIZATION}" \
|
||||
--advantage-clip "${ADVANTAGE_CLIP}" \
|
||||
"${PPO_CLIP_ARGS[@]}" \
|
||||
--ppo-clip-low "${PPO_CLIP_LOW}" \
|
||||
--ppo-clip-high "${PPO_CLIP_HIGH}" \
|
||||
--sampled-loss-reduction "${SAMPLED_LOSS_REDUCTION}" \
|
||||
"${POS_ADV_ARGS[@]}" \
|
||||
--max-length "${MAX_LENGTH}" \
|
||||
--distill-chunk-size "${DISTILL_CHUNK_SIZE}" \
|
||||
--per-device-train-batch-size "${PER_DEVICE_TRAIN_BATCH_SIZE}" \
|
||||
--gradient-accumulation-steps "${GRADIENT_ACCUMULATION_STEPS}" \
|
||||
--learning-rate "${LEARNING_RATE}" \
|
||||
--weight-decay "${WEIGHT_DECAY}" \
|
||||
--adam-beta1 "${ADAM_BETA1}" \
|
||||
--adam-beta2 "${ADAM_BETA2}" \
|
||||
--warmup-ratio "${WARMUP_RATIO}" \
|
||||
--lr-scheduler-type "${LR_SCHEDULER_TYPE}" \
|
||||
--max-steps "${MAX_STEPS}" \
|
||||
--save-steps "${SAVE_STEPS}" \
|
||||
--logging-steps "${LOGGING_STEPS}" \
|
||||
--freeze-lora-b-after-step "${FREEZE_LORA_B_AFTER_STEP}" \
|
||||
--no-gradient-checkpointing \
|
||||
"$@"
|
||||
19
scripts/serve_teacher_32b.sh
Normal file
19
scripts/serve_teacher_32b.sh
Normal file
@@ -0,0 +1,19 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
LOG_FILE="/tmp/sglang_$(head /dev/urandom | tr -dc A-Za-z0-9 | head -c 6).log"
|
||||
python3 -m sglang.launch_server \
|
||||
--model-path Qwen/Qwen3-32B \
|
||||
--host 127.0.0.1 \
|
||||
--port 13141 \
|
||||
--tp 8 \
|
||||
--chunked-prefill-size 4096 \
|
||||
--mem-fraction-static 0.6 \
|
||||
--context-length 8192 \
|
||||
> "$LOG_FILE" 2>&1 &
|
||||
|
||||
until curl -sf http://127.0.0.1:13141/health_generate > /dev/null; do
|
||||
echo "Waiting for the teacher model server to start..."
|
||||
tail -n 10 "$LOG_FILE"
|
||||
sleep 5
|
||||
done
|
||||
19
scripts/serve_teacher_8b.sh
Normal file
19
scripts/serve_teacher_8b.sh
Normal file
@@ -0,0 +1,19 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
LOG_FILE="/tmp/sglang_$(head /dev/urandom | tr -dc A-Za-z0-9 | head -c 6).log"
|
||||
python3 -m sglang.launch_server \
|
||||
--model-path model_weights/qwen3-8b \
|
||||
--host 127.0.0.1 \
|
||||
--port 13141 \
|
||||
--tp 4 \
|
||||
--chunked-prefill-size 4096 \
|
||||
--mem-fraction-static 0.6 \
|
||||
--context-length 8192 \
|
||||
> "$LOG_FILE" 2>&1 &
|
||||
|
||||
until curl -sf http://127.0.0.1:13141/health_generate > /dev/null; do
|
||||
echo "Waiting for the teacher model server to start..."
|
||||
tail -n 10 "$LOG_FILE"
|
||||
sleep 5
|
||||
done
|
||||
57
setup.py
Normal file
57
setup.py
Normal file
@@ -0,0 +1,57 @@
|
||||
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
|
||||
import sys
|
||||
import platform
|
||||
|
||||
from setuptools import find_packages, setup
|
||||
from wheel.bdist_wheel import bdist_wheel as _bdist_wheel
|
||||
|
||||
|
||||
def _fetch_requirements(path):
|
||||
with open(path, "r") as fd:
|
||||
return [r.strip() for r in fd.readlines() if r.strip() and not r.startswith("#")]
|
||||
|
||||
|
||||
# Custom wheel class to modify the wheel name
|
||||
class bdist_wheel(_bdist_wheel):
|
||||
def finalize_options(self):
|
||||
_bdist_wheel.finalize_options(self)
|
||||
self.root_is_pure = False
|
||||
|
||||
def get_tag(self):
|
||||
python_version = f"cp{sys.version_info.major}{sys.version_info.minor}"
|
||||
abi_tag = f"{python_version}"
|
||||
|
||||
if platform.system() == "Linux":
|
||||
platform_tag = "manylinux1_x86_64"
|
||||
else:
|
||||
platform_tag = platform.system().lower()
|
||||
|
||||
return python_version, abi_tag, platform_tag
|
||||
|
||||
|
||||
# Setup configuration
|
||||
setup(
|
||||
author="slime Team",
|
||||
name="slime",
|
||||
version="0.1.0",
|
||||
packages=find_packages(include=["slime*", "slime_plugins*"]),
|
||||
include_package_data=True,
|
||||
install_requires=_fetch_requirements("requirements.txt"),
|
||||
extras_require={
|
||||
"fsdp": [
|
||||
"torch>=2.0",
|
||||
]
|
||||
},
|
||||
python_requires=">=3.10",
|
||||
classifiers=[
|
||||
"Programming Language :: Python :: 3.10",
|
||||
"Programming Language :: Python :: 3.11",
|
||||
"Programming Language :: Python :: 3.12",
|
||||
"Environment :: GPU :: NVIDIA CUDA",
|
||||
"Topic :: Scientific/Engineering :: Artificial Intelligence",
|
||||
"Topic :: System :: Distributed Computing",
|
||||
],
|
||||
cmdclass={"bdist_wheel": bdist_wheel},
|
||||
)
|
||||
39
slime.egg-info/PKG-INFO
Normal file
39
slime.egg-info/PKG-INFO
Normal file
@@ -0,0 +1,39 @@
|
||||
Metadata-Version: 2.4
|
||||
Name: slime
|
||||
Version: 0.1.0
|
||||
Author: slime Team
|
||||
Classifier: Programming Language :: Python :: 3.10
|
||||
Classifier: Programming Language :: Python :: 3.11
|
||||
Classifier: Programming Language :: Python :: 3.12
|
||||
Classifier: Environment :: GPU :: NVIDIA CUDA
|
||||
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
|
||||
Classifier: Topic :: System :: Distributed Computing
|
||||
Requires-Python: >=3.10
|
||||
License-File: LICENSE
|
||||
Requires-Dist: accelerate
|
||||
Requires-Dist: blobfile
|
||||
Requires-Dist: datasets
|
||||
Requires-Dist: httpx[http2]
|
||||
Requires-Dist: mcp[cli]
|
||||
Requires-Dist: megatron-bridge @ git+https://github.com/fzyzcjy/Megatron-Bridge.git@dev_rl
|
||||
Requires-Dist: memray
|
||||
Requires-Dist: nvidia-modelopt[torch]>=0.37.0
|
||||
Requires-Dist: omegaconf
|
||||
Requires-Dist: pillow
|
||||
Requires-Dist: pylatexenc
|
||||
Requires-Dist: pyyaml
|
||||
Requires-Dist: ray[default]
|
||||
Requires-Dist: ring_flash_attn
|
||||
Requires-Dist: sglang-router>=0.2.3
|
||||
Requires-Dist: tensorboard
|
||||
Requires-Dist: transformers
|
||||
Requires-Dist: wandb
|
||||
Requires-Dist: liger_kernel
|
||||
Provides-Extra: fsdp
|
||||
Requires-Dist: torch>=2.0; extra == "fsdp"
|
||||
Dynamic: author
|
||||
Dynamic: classifier
|
||||
Dynamic: license-file
|
||||
Dynamic: provides-extra
|
||||
Dynamic: requires-dist
|
||||
Dynamic: requires-python
|
||||
140
slime.egg-info/SOURCES.txt
Normal file
140
slime.egg-info/SOURCES.txt
Normal file
@@ -0,0 +1,140 @@
|
||||
LICENSE
|
||||
README.md
|
||||
pyproject.toml
|
||||
setup.py
|
||||
slime/__init__.py
|
||||
slime.egg-info/PKG-INFO
|
||||
slime.egg-info/SOURCES.txt
|
||||
slime.egg-info/dependency_links.txt
|
||||
slime.egg-info/requires.txt
|
||||
slime.egg-info/top_level.txt
|
||||
slime/backends/__init__.py
|
||||
slime/backends/fsdp_utils/__init__.py
|
||||
slime/backends/fsdp_utils/actor.py
|
||||
slime/backends/fsdp_utils/arguments.py
|
||||
slime/backends/fsdp_utils/checkpoint.py
|
||||
slime/backends/fsdp_utils/data_packing.py
|
||||
slime/backends/fsdp_utils/lr_scheduler.py
|
||||
slime/backends/fsdp_utils/update_weight_utils.py
|
||||
slime/backends/fsdp_utils/kernels/__init__.py
|
||||
slime/backends/fsdp_utils/kernels/fused_experts.py
|
||||
slime/backends/fsdp_utils/kernels/fused_moe_triton_backward_kernels.py
|
||||
slime/backends/fsdp_utils/models/__init__.py
|
||||
slime/backends/fsdp_utils/models/qwen3_moe.py
|
||||
slime/backends/fsdp_utils/models/qwen3_moe_hf.py
|
||||
slime/backends/megatron_utils/__init__.py
|
||||
slime/backends/megatron_utils/actor.py
|
||||
slime/backends/megatron_utils/arguments.py
|
||||
slime/backends/megatron_utils/checkpoint.py
|
||||
slime/backends/megatron_utils/ci_utils.py
|
||||
slime/backends/megatron_utils/cp_utils.py
|
||||
slime/backends/megatron_utils/data.py
|
||||
slime/backends/megatron_utils/initialize.py
|
||||
slime/backends/megatron_utils/loss.py
|
||||
slime/backends/megatron_utils/misc_utils.py
|
||||
slime/backends/megatron_utils/model.py
|
||||
slime/backends/megatron_utils/model_provider.py
|
||||
slime/backends/megatron_utils/sglang.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/__init__.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/deepseekv3.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/glm4.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/glm4moe.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/llama.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/mimo.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/qwen2.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/qwen3_next.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/qwen3moe.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/processors/__init__.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/processors/padding_remover.py
|
||||
slime/backends/megatron_utils/megatron_to_hf/processors/quantizer.py
|
||||
slime/backends/megatron_utils/update_weight/__init__.py
|
||||
slime/backends/megatron_utils/update_weight/common.py
|
||||
slime/backends/megatron_utils/update_weight/hf_weight_iterator_base.py
|
||||
slime/backends/megatron_utils/update_weight/hf_weight_iterator_bridge.py
|
||||
slime/backends/megatron_utils/update_weight/hf_weight_iterator_direct.py
|
||||
slime/backends/megatron_utils/update_weight/update_weight_from_distributed.py
|
||||
slime/backends/megatron_utils/update_weight/update_weight_from_tensor.py
|
||||
slime/backends/sglang_utils/__init__.py
|
||||
slime/backends/sglang_utils/arguments.py
|
||||
slime/backends/sglang_utils/sglang_engine.py
|
||||
slime/ray/__init__.py
|
||||
slime/ray/actor_group.py
|
||||
slime/ray/placement_group.py
|
||||
slime/ray/ray_actor.py
|
||||
slime/ray/rollout.py
|
||||
slime/ray/train_actor.py
|
||||
slime/ray/utils.py
|
||||
slime/rollout/__init__.py
|
||||
slime/rollout/base_types.py
|
||||
slime/rollout/data_source.py
|
||||
slime/rollout/on_policy_distillation.py
|
||||
slime/rollout/sglang_rollout.py
|
||||
slime/rollout/sleep_rollout.py
|
||||
slime/rollout/filter_hub/__init__.py
|
||||
slime/rollout/filter_hub/base_types.py
|
||||
slime/rollout/filter_hub/dynamic_sampling_filters.py
|
||||
slime/rollout/rm_hub/__init__.py
|
||||
slime/rollout/rm_hub/deepscaler.py
|
||||
slime/rollout/rm_hub/f1.py
|
||||
slime/rollout/rm_hub/gpqa.py
|
||||
slime/rollout/rm_hub/ifbench.py
|
||||
slime/rollout/rm_hub/math_dapo_utils.py
|
||||
slime/rollout/rm_hub/math_utils.py
|
||||
slime/router/__init__.py
|
||||
slime/router/router.py
|
||||
slime/router/middleware_hub/__init__.py
|
||||
slime/router/middleware_hub/radix_tree.py
|
||||
slime/router/middleware_hub/radix_tree_middleware.py
|
||||
slime/utils/__init__.py
|
||||
slime/utils/arguments.py
|
||||
slime/utils/async_utils.py
|
||||
slime/utils/context_utils.py
|
||||
slime/utils/data.py
|
||||
slime/utils/distributed_utils.py
|
||||
slime/utils/eval_config.py
|
||||
slime/utils/flops_utils.py
|
||||
slime/utils/fp8_kernel.py
|
||||
slime/utils/health_monitor.py
|
||||
slime/utils/http_utils.py
|
||||
slime/utils/iter_utils.py
|
||||
slime/utils/logging_utils.py
|
||||
slime/utils/mask_utils.py
|
||||
slime/utils/megatron_bridge_utils.py
|
||||
slime/utils/memory_utils.py
|
||||
slime/utils/metric_checker.py
|
||||
slime/utils/metric_utils.py
|
||||
slime/utils/misc.py
|
||||
slime/utils/ppo_utils.py
|
||||
slime/utils/processing_utils.py
|
||||
slime/utils/profile_utils.py
|
||||
slime/utils/ray_utils.py
|
||||
slime/utils/reloadable_process_group.py
|
||||
slime/utils/rocm_checkpoint_writer.py
|
||||
slime/utils/routing_replay.py
|
||||
slime/utils/seqlen_balancing.py
|
||||
slime/utils/tensor_backper.py
|
||||
slime/utils/tensorboard_utils.py
|
||||
slime/utils/timer.py
|
||||
slime/utils/tracking_utils.py
|
||||
slime/utils/train_dump_utils.py
|
||||
slime/utils/train_metric_utils.py
|
||||
slime/utils/typer_utils.py
|
||||
slime/utils/types.py
|
||||
slime/utils/wandb_utils.py
|
||||
slime/utils/debug_utils/__init__.py
|
||||
slime/utils/debug_utils/display_debug_rollout_data.py
|
||||
slime/utils/debug_utils/replay_reward_fn.py
|
||||
slime/utils/debug_utils/send_to_sglang.py
|
||||
slime/utils/external_utils/__init__.py
|
||||
slime/utils/external_utils/command_utils.py
|
||||
slime_plugins/__init__.py
|
||||
slime_plugins/mbridge/__init__.py
|
||||
slime_plugins/mbridge/glm4.py
|
||||
slime_plugins/mbridge/glm4moe.py
|
||||
slime_plugins/mbridge/mimo.py
|
||||
slime_plugins/mbridge/qwen3_next.py
|
||||
slime_plugins/megatron_bridge/__init__.py
|
||||
slime_plugins/models/__init__.py
|
||||
slime_plugins/models/glm4.py
|
||||
slime_plugins/models/hf_attention.py
|
||||
slime_plugins/models/qwen3_next.py
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user