Shanghai_AI_Laboratory/internlm2-math-20b
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

upload internlm2-math-20b

Browse Source
This commit is contained in:
x54-729
2024-01-26 13:14:07 +00:00
parent 0af374e301
commit dc21129ed8
33 changed files with 2691 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
.gitattributes vendored
View File

@@ -31,4 +31,25 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
**/*ckpt*.meta filter=lfs diff=lfs merge=lfs -text
**/*ckpt*.index filter=lfs diff=lfs merge=lfs -text
*.safetensors filter=lfs diff=lfs merge=lfs -text
*.ckpt filter=lfs diff=lfs merge=lfs -text
*.ckpt filter=lfs diff=lfs merge=lfs -text
model-00001-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00003-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00005-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00006-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00014-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00004-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00011-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00012-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00017-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00020-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00002-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00009-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00015-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00021-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00007-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00008-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00010-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00013-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00016-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00018-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text
model-00019-of-00021.safetensors filter=lfs diff=lfs merge=lfs -text

142
README.md
View File

@@ -1,13 +1,133 @@
---
frameworks:
- Pytorch
license: Apache License 2.0
tasks:
- text-generation
pipeline_tag: text-generation
license: other
language:
- en
- zh
tags:
- math
---
###### 该模型当前使用的是默认介绍模版，处于“预发布”阶段，页面仅限所有者可见。
###### 请根据[模型贡献文档说明](https://www.modelscope.cn/docs/%E5%A6%82%E4%BD%95%E6%92%B0%E5%86%99%E5%A5%BD%E7%94%A8%E7%9A%84%E6%A8%A1%E5%9E%8B%E5%8D%A1%E7%89%87)，及时完善模型卡片内容。ModelScope平台将在模型卡片完善后展示。谢谢您的理解。
#### Clone with HTTP
```bash
git clone https://www.modelscope.cn/Shanghai_AI_Laboratory/internlm2-math-20b.git
```
# InternLM-Math
<div align="center">
<img src="https://raw.githubusercontent.com/InternLM/InternLM/main/assets/logo.svg" width="200"/>
<div> </div>
<div align="center">
<b><font size="5">InternLM-Math</font></b>
<sup>
<a href="https://internlm.intern-ai.org.cn/">
<i><font size="4">HOT</font></i>
</a>
</sup>
<div> </div>
</div>
State-of-the-art bilingual open-sourced Math reasoning LLMs.
</div>
# Introduction
- **7B and 20B Chinese and English Math LMs with better than ChatGPT performances.** InternLM2-Math are continued pretrained from InternLM2-Base with ~100B high quality math-related tokens and SFT with ~2M bilingual math supervised data. We apply minhash and exact number match to decontaminate possible test set leakage.
- **Add Lean as a support language for math problem solving and math theorem proving.** We are exploring combining Lean 3 with InternLM-Math for verifiable math reasoning. InternLM-Math can generate Lean codes for simple math reasoning tasks like GSM8K or provide possible proof tactics based on Lean states.
- **Also can be viewed as a reward model, which supports the Outcome/Process/Lean Reward Model.** We supervise InternLM2-Math with various types of reward modeling data, to make InternLM2-Math can also verify chain-of-thought processes. We also add the ability to convert a chain-of-thought process into Lean 3 code.
- **A Math LM Augment Helper** and **Code Intepreter**. InternLM2-Math can help augment math reasoning problems and solve them using the code interpreter which makes you generate synthesis data quicker!
# Models
| Model | Transformers(HF) |Release Date |
|---|---|---|
| **InternLM2-Math-Base-7B** | [🤗internlm/internlm2-math-base-7b](https://huggingface.co/internlm/internlm2-math-base-7b) | 2024-01-23|
| **InternLM2-Math-Base-20B** | [🤗internlm/internlm2-math-base-20b](https://huggingface.co/internlm/internlm2-math-base-20b) | 2024-01-23|
| **InternLM2-Math-7B** | [🤗internlm/internlm2-math-7b](https://huggingface.co/internlm/internlm2-math-7b) | 2024-01-23|
| **InternLM2-Math-20B** | [🤗internlm/internlm2-math-20b](https://huggingface.co/internlm/internlm2-math-20b) | 2024-01-23|
# Performance
## Pretrain Performance
We evaluate pretrain checkpoints based on greedy decoding with few-shot COT. Details of pretraining will be introduced in the tech report.
| Model | GSM8K | MATH |
|------------------------|---------|--------|
| Llama2-7B | 11.8 | 3.2 |
| Llemma-7B | 36.4 | 18.0 |
| InternLM2-Base-7B | 36.5 | 8.6 |
| **InternLM2-Math-Base-7B** | **49.2** | **21.5** |
| Minerva-8B | 16.2 | 14.1 |
| InternLM2-Base-20B | 54.6 | 13.7 |
| **InternLM2-Math-Base-20B** | **63.7** | **27.3** |
| Llemma-34B | 51.5 | 25.0 |
| Minerva-62B | 52.4 | 27.6 |
| Minerva-540B | 58.8 | 33.6 |
## SFT Peformance
All performance is based on greedy decoding with COT. We notice that the performance of Hungary has a big variance between our different checkpoints, while other performance is very stable. This may be due to the problem amount about Hungary.
| Model | Model Type | GSM8K | MATH | Hungary |
|------------------------|----------------------|--------|--------|---------|
| Qwen-7B-Chat | Genearl | 51.7 | 11.6 | - |
| DeepSeek-7B-Chat | General | 63.0 | 15.8 | 28.5 |
| InternLM2-Chat-7B | General | 70.7 | 23.0 | - |
| ChatGLM3-6B | General | 53.8 | 20.4 | 32 |
| MetaMath-Mistral-7B | Mathematics | 77.7 | 28.2 | 29 |
| MetaMath-Llemma-7B | Mathematics | 69.2 | 30.0 | - |
| **InternLM2-Math-7B** | Mathematics | **78.1** | **34.6** | **55** |
| InternLM2-Chat-20B | General | 79.6 | 31.9 | - |
| MetaMath-Llemma-34B | Mathematics | 75.8 | 34.8 | - |
| **InternLM2-Math-20B** | Mathematics | **82.6** | **37.7** | **66** |
| Qwen-72B | General | 78.9 | 35.2 | 52 |
| DeepSeek-67B | General | 84.1 | 32.6 | 58 |
| ChatGPT (GPT-3.5) | General | 80.8 | 34.1 | 41 |
| GPT4 (First version) | General | 92.0 | 42.5 | 68 |
# Inference
```python
from modelscope import snapshot_download, AutoTokenizer, AutoModelForCausalLM
import torch
model_dir = snapshot_download("Shanghai_AI_Laboratory/internlm2-math-20b")
tokenizer = AutoTokenizer.from_pretrained(model_dir, device_map="auto", trust_remote_code=True)
# Set `torch_dtype=torch.float16` to load model in float16, otherwise it will be loaded as float32 and might cause OOM Error.
model = AutoModelForCausalLM.from_pretrained(model_dir, device_map="auto", trust_remote_code=True, torch_dtype=torch.float16)
model = model.eval()
response, history = model.chat(tokenizer, "1+1=", history=[], meta_instruction="")
print(response)
```
# Special usages
We list some instructions used in our SFT. You can use them to help you. You can use the other ways to prompt the model, but the following are recommended. InternLM2-Math may combine the following abilities but it is not guaranteed.
| Description | Query |
| --- | --- |
| Solving question via chain-of-thought | {Question} |
| Solving question via Lean 3 | {Question}\nSolve this via Lean 3 |
| Outcome reward model | Given a question and an answer, check is it correct?\nQuestion:{Question}\nAnswer:{COT} |
| Process reward model | Given a question and an answer, check correctness of each step.\nQuestion:{Question}\nAnswer:{COT} |
| Reward model | Given a question and two answers, which one is better? \nQuestion:{Question}\nAnswer 1:{COT}\nAnswer 2:{COT} |
| Convert chain-of-thought to Lean 3 | Convert this answer into Lean3. Question:{Question}\nAnswer:{COT} |
| Convert Lean 3 to chain-of-thought | Convert this lean 3 code into a natural language problem with answers:\n{LEAN} |
| Translate question and chain-of-thought answer to a proof statement | Convert this question and answer into a proof format.\nQuestion:{Question}\nAnswer:{COT} |
| Translate proof problem to Lean 3 | Convert this natural langauge statement into a Lean 3 theorem statement:{Theorem} |
| Translate Lean 3 to proof problem | Convert this Lean 3 theorem statement into natural language:{STATEMENT} |
| Suggest a tactic based on Lean state | Given the Lean 3 tactic state, suggest a next tactic:\n{State} |
| Rephrase Problem | Describe this problem in another way. {STATEMENT} |
| Augment Problem | Please augment a new problem based on: {Question} |
| Augment a harder Problem | Increase the complexity of the problem: {Question} |
| Change specific numbers | Change specific numbers: {Question}|
| Introduce fractions or percentages | Introduce fractions or percentages: {Question}|
| Code Intepreter | [lagent](https://github.com/InternLM/InternLM/blob/main/agent/lagent.md) |
| In-context Learning | Question:{Question}\nAnswer:{COT}\n...Question:{Question}\nAnswer:{COT}|
# Fine-tune and others
Please refer to [InternLM](https://github.com/InternLM/InternLM/tree/main).
# Known issues
Our model is still under development and will be upgraded. There are some possible issues of InternLM-Math.
- Jump the calculating step.
- Perform badly at Chinese fill-in-the-bank problems and English choice problems due to SFT data composition.
- The reward model mode can be better leveraged with assigned token probabilities.
- Code switch due to SFT data composition.
- Some abilities of Lean can only be adapted to GSM8K-like problems (e.g. Convert chain-of-thought to Lean 3), and performance related to Lean is not guaranteed.
# Citation and Tech Report
To be appended.

31
config.json Normal file
View File

@@ -0,0 +1,31 @@
{
"architectures": [
"InternLM2ForCausalLM"
],
"auto_map": {
"AutoConfig": "configuration_internlm2.InternLM2Config",
"AutoModelForCausalLM": "modeling_internlm2.InternLM2ForCausalLM",
"AutoModel": "modeling_internlm2.InternLM2ForCausalLM"
},
"bias": false,
"bos_token_id": 1,
"eos_token_id": 2,
"hidden_act": "silu",
"hidden_size": 6144,
"initializer_range": 0.02,
"intermediate_size": 16384,
"max_position_embeddings": 8192,
"model_type": "internlm2",
"num_attention_heads": 48,
"num_hidden_layers": 48,
"num_key_value_heads": 8,
"pad_token_id": 2,
"rms_norm_eps": 1e-05,
"rope_scaling": null,
"rope_theta": 1000000,
"tie_word_embeddings": false,
"torch_dtype": "bfloat16",
"transformers_version": "4.35.2",
"use_cache": true,
"vocab_size": 92544
}

151
configuration_internlm2.py Normal file
View File

@@ -0,0 +1,151 @@
# coding=utf-8
# Copyright (c) The InternLM team and The HuggingFace Inc. team. All rights reserved.
#
# This code is based on transformers/src/transformers/models/llama/configuration_llama.py
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
""" InternLM2 model configuration"""
from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging
logger = logging.get_logger(__name__)
INTERNLM2_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
# Modified from transformers.model.llama.configuration_llama.LlamaConfig
class InternLM2Config(PretrainedConfig):
r"""
This is the configuration class to store the configuration of a [`InternLM2Model`]. It is used to instantiate
an InternLM2 model according to the specified arguments, defining the model architecture. Instantiating a
configuration with the defaults will yield a similar configuration to that of the InternLM2-7B.
Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
documentation from [`PretrainedConfig`] for more information.
Args:
vocab_size (`int`, *optional*, defaults to 32000):
Vocabulary size of the InternLM2 model. Defines the number of different tokens that can be represented by the
`inputs_ids` passed when calling [`InternLM2Model`]
hidden_size (`int`, *optional*, defaults to 4096):
Dimension of the hidden representations.
intermediate_size (`int`, *optional*, defaults to 11008):
Dimension of the MLP representations.
num_hidden_layers (`int`, *optional*, defaults to 32):
Number of hidden layers in the Transformer encoder.
num_attention_heads (`int`, *optional*, defaults to 32):
Number of attention heads for each attention layer in the Transformer encoder.
num_key_value_heads (`int`, *optional*):
This is the number of key_value heads that should be used to implement Grouped Query Attention. If
`num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
`num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
by meanpooling all the original heads within that group. For more details checkout [this
paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
`num_attention_heads`.
hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
The non-linear activation function (function or string) in the decoder.
max_position_embeddings (`int`, *optional*, defaults to 2048):
The maximum sequence length that this model might ever be used with. Typically set this to something large
just in case (e.g., 512 or 1024 or 2048).
initializer_range (`float`, *optional*, defaults to 0.02):
The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
rms_norm_eps (`float`, *optional*, defaults to 1e-12):
The epsilon used by the rms normalization layers.
use_cache (`bool`, *optional*, defaults to `True`):
Whether or not the model should return the last key/values attentions (not used by all models). Only
relevant if `config.is_decoder=True`.
tie_word_embeddings(`bool`, *optional*, defaults to `False`):
Whether to tie weight embeddings
Example:
"""
model_type = "internlm2"
_auto_class = "AutoConfig"
def __init__( # pylint: disable=W0102
self,
vocab_size=103168,
hidden_size=4096,
intermediate_size=11008,
num_hidden_layers=32,
num_attention_heads=32,
num_key_value_heads=None,
hidden_act="silu",
max_position_embeddings=2048,
initializer_range=0.02,
rms_norm_eps=1e-6,
use_cache=True,
pad_token_id=0,
bos_token_id=1,
eos_token_id=2,
tie_word_embeddings=False,
bias=True,
rope_theta=10000,
rope_scaling=None,
attn_implementation="eager",
**kwargs,
):
self.vocab_size = vocab_size
self.max_position_embeddings = max_position_embeddings
self.hidden_size = hidden_size
self.intermediate_size = intermediate_size
self.num_hidden_layers = num_hidden_layers
self.num_attention_heads = num_attention_heads
self.bias = bias
if num_key_value_heads is None:
num_key_value_heads = num_attention_heads
self.num_key_value_heads = num_key_value_heads
self.hidden_act = hidden_act
self.initializer_range = initializer_range
self.rms_norm_eps = rms_norm_eps
self.use_cache = use_cache
self.rope_theta = rope_theta
self.rope_scaling = rope_scaling
self._rope_scaling_validation()
self.attn_implementation = attn_implementation
if self.attn_implementation is None:
self.attn_implementation = "eager"
super().__init__(
pad_token_id=pad_token_id,
bos_token_id=bos_token_id,
eos_token_id=eos_token_id,
tie_word_embeddings=tie_word_embeddings,
**kwargs,
)
def _rope_scaling_validation(self):
"""
Validate the `rope_scaling` configuration.
"""
if self.rope_scaling is None:
return
if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
raise ValueError(
"`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
f"got {self.rope_scaling}"
)
rope_scaling_type = self.rope_scaling.get("type", None)
rope_scaling_factor = self.rope_scaling.get("factor", None)
if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
raise ValueError(
f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
)
if rope_scaling_factor is None or not isinstance(rope_scaling_factor, float) or rope_scaling_factor < 1.0:
raise ValueError(f"`rope_scaling`'s factor field must be a float >= 1, got {rope_scaling_factor}")

7
generation_config.json Normal file
View File

@@ -0,0 +1,7 @@
{
"_from_model_config": true,
"bos_token_id": 1,
"eos_token_id": 2,
"pad_token_id": 2,
"transformers_version": "4.35.2"
}

3
model-00001-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:2d76ba004a3e3ccfc3b58e3178fe901a8e62511443ece2896209bd8e0f36b6b2
size 1917346712

3
model-00002-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:15a67775bc6960cce5c229886f4654fd7c3d7179a6777011c51d81fecf58ee14
size 1937819544

3
model-00003-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cd742b765847b6cd2df9271aba2be3b20bfb2d49a598b7d0ecb131cb13a1cc41
size 1963010040

3
model-00004-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:525a26e9e521996f8ed8bd3671def0b3187a3ea64bc71ce160cabec53ba22d70
size 1937819544

3
model-00005-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:4bedbbf363ac1893331fa49b397b7434bd107a20239f108c3680368455d4195c
size 1963010056

3
model-00006-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f0f7409abb93eae0c9610b40b1d0ca1c63c3e268f7b763929069a729a2c89ab2
size 1937819560

3
model-00007-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c0fe2222b25fde896ff8f7e49755b0cd53966e9f2a8b69560c8995dbe4fa0e6e
size 1963010064

3
model-00008-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:484f476615b282d00cd03765f6db910b202397ae69e3fe76e3af966cb99b165a
size 1937819560

3
model-00009-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:530addf4b3600bc47b4f0c9214a45ca3256610e809fb17b6311521b726a41ec4
size 1963010064

3
model-00010-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:c386a84da17e9e6dc7e64858f6d8678bff6cd9b97bac43704551669e6d821556
size 1937819560

3
model-00011-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:b3f4370c7ad210df0f4775f64e9e0cf1daceaf1462efd77ff2d345b87e685b31
size 1963010064

3
model-00012-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:9fa0e34c1a8a6cd79f796965fb53a629bddc4f25bc8693cacfe57d245a7ecaf5
size 1937819560

3
model-00013-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:3d759ebd095a2a4d6be4d5970c78902375cdde1c23cc5d10b9646f23b03afde0
size 1963010064

3
model-00014-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cb3847692a87d01d18f4bc30c3699bfdcc61df762bc91967cffb291eaa671f3c
size 1937819560

3
model-00015-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:09eca575963b7ae7d26d79a0a81626f7ddb5f9cf9e66f4c423929d75e277c189
size 1963010064

3
model-00016-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f013c9cb10fc939a7e6dec0f2b8136f46dedfce3a9cb0faf7bc1e9a404a2d41a
size 1937819560

3
model-00017-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:580de14cb160de4fb039f17499c8b24a15ace432b69b8991d7788d5d3c481f0a
size 1963010064

3
model-00018-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:cb25638a8d82192385598ed6efcf386a090d7ef2d5c9e33e84c80210403a65ac
size 1937819560

3
model-00019-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d2fbf8eb25674346cab5ecd31d3d85abe46722c895e629983f2d0c1fcbee1479
size 1963010064

3
model-00020-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:d6ec2fe0740539dc2caf905262e965098aa35f0dc1784e3ef90059371a02ca8f
size 1560344232

3
model-00021-of-00021.safetensors Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:e1b8e046517b8e5be8cd526fd48fc25f7c33add377d5c0735cf29fae116a2ceb
size 1137180800

346
model.safetensors.index.json Normal file
View File

@@ -0,0 +1,346 @@
{
"metadata": {
"total_size": 39722299392
},
"weight_map": {
"model.layers.0.attention.wo.weight": "model-00001-of-00021.safetensors",
"model.layers.0.attention.wqkv.weight": "model-00001-of-00021.safetensors",
"model.layers.0.attention_norm.weight": "model-00001-of-00021.safetensors",
"model.layers.0.feed_forward.w1.weight": "model-00001-of-00021.safetensors",
"model.layers.0.feed_forward.w2.weight": "model-00001-of-00021.safetensors",
"model.layers.0.feed_forward.w3.weight": "model-00001-of-00021.safetensors",
"model.layers.0.ffn_norm.weight": "model-00001-of-00021.safetensors",
"model.layers.1.attention.wo.weight": "model-00002-of-00021.safetensors",
"model.layers.1.attention.wqkv.weight": "model-00002-of-00021.safetensors",
"model.layers.1.attention_norm.weight": "model-00002-of-00021.safetensors",
"model.layers.1.feed_forward.w1.weight": "model-00002-of-00021.safetensors",
"model.layers.1.feed_forward.w2.weight": "model-00002-of-00021.safetensors",
"model.layers.1.feed_forward.w3.weight": "model-00002-of-00021.safetensors",
"model.layers.1.ffn_norm.weight": "model-00002-of-00021.safetensors",
"model.layers.10.attention.wo.weight": "model-00005-of-00021.safetensors",
"model.layers.10.attention.wqkv.weight": "model-00005-of-00021.safetensors",
"model.layers.10.attention_norm.weight": "model-00005-of-00021.safetensors",
"model.layers.10.feed_forward.w1.weight": "model-00005-of-00021.safetensors",
"model.layers.10.feed_forward.w2.weight": "model-00005-of-00021.safetensors",
"model.layers.10.feed_forward.w3.weight": "model-00005-of-00021.safetensors",
"model.layers.10.ffn_norm.weight": "model-00005-of-00021.safetensors",
"model.layers.11.attention.wo.weight": "model-00006-of-00021.safetensors",
"model.layers.11.attention.wqkv.weight": "model-00006-of-00021.safetensors",
"model.layers.11.attention_norm.weight": "model-00006-of-00021.safetensors",
"model.layers.11.feed_forward.w1.weight": "model-00006-of-00021.safetensors",
"model.layers.11.feed_forward.w2.weight": "model-00006-of-00021.safetensors",
"model.layers.11.feed_forward.w3.weight": "model-00006-of-00021.safetensors",
"model.layers.11.ffn_norm.weight": "model-00006-of-00021.safetensors",
"model.layers.12.attention.wo.weight": "model-00006-of-00021.safetensors",
"model.layers.12.attention.wqkv.weight": "model-00006-of-00021.safetensors",
"model.layers.12.attention_norm.weight": "model-00006-of-00021.safetensors",
"model.layers.12.feed_forward.w1.weight": "model-00006-of-00021.safetensors",
"model.layers.12.feed_forward.w2.weight": "model-00006-of-00021.safetensors",
"model.layers.12.feed_forward.w3.weight": "model-00006-of-00021.safetensors",
"model.layers.12.ffn_norm.weight": "model-00006-of-00021.safetensors",
"model.layers.13.attention.wo.weight": "model-00006-of-00021.safetensors",
"model.layers.13.attention.wqkv.weight": "model-00006-of-00021.safetensors",
"model.layers.13.attention_norm.weight": "model-00007-of-00021.safetensors",
"model.layers.13.feed_forward.w1.weight": "model-00006-of-00021.safetensors",
"model.layers.13.feed_forward.w2.weight": "model-00007-of-00021.safetensors",
"model.layers.13.feed_forward.w3.weight": "model-00007-of-00021.safetensors",
"model.layers.13.ffn_norm.weight": "model-00007-of-00021.safetensors",
"model.layers.14.attention.wo.weight": "model-00007-of-00021.safetensors",
"model.layers.14.attention.wqkv.weight": "model-00007-of-00021.safetensors",
"model.layers.14.attention_norm.weight": "model-00007-of-00021.safetensors",
"model.layers.14.feed_forward.w1.weight": "model-00007-of-00021.safetensors",
"model.layers.14.feed_forward.w2.weight": "model-00007-of-00021.safetensors",
"model.layers.14.feed_forward.w3.weight": "model-00007-of-00021.safetensors",
"model.layers.14.ffn_norm.weight": "model-00007-of-00021.safetensors",
"model.layers.15.attention.wo.weight": "model-00007-of-00021.safetensors",
"model.layers.15.attention.wqkv.weight": "model-00007-of-00021.safetensors",
"model.layers.15.attention_norm.weight": "model-00007-of-00021.safetensors",
"model.layers.15.feed_forward.w1.weight": "model-00007-of-00021.safetensors",
"model.layers.15.feed_forward.w2.weight": "model-00007-of-00021.safetensors",
"model.layers.15.feed_forward.w3.weight": "model-00007-of-00021.safetensors",
"model.layers.15.ffn_norm.weight": "model-00007-of-00021.safetensors",
"model.layers.16.attention.wo.weight": "model-00008-of-00021.safetensors",
"model.layers.16.attention.wqkv.weight": "model-00008-of-00021.safetensors",
"model.layers.16.attention_norm.weight": "model-00008-of-00021.safetensors",
"model.layers.16.feed_forward.w1.weight": "model-00008-of-00021.safetensors",
"model.layers.16.feed_forward.w2.weight": "model-00008-of-00021.safetensors",
"model.layers.16.feed_forward.w3.weight": "model-00008-of-00021.safetensors",
"model.layers.16.ffn_norm.weight": "model-00008-of-00021.safetensors",
"model.layers.17.attention.wo.weight": "model-00008-of-00021.safetensors",
"model.layers.17.attention.wqkv.weight": "model-00008-of-00021.safetensors",
"model.layers.17.attention_norm.weight": "model-00008-of-00021.safetensors",
"model.layers.17.feed_forward.w1.weight": "model-00008-of-00021.safetensors",
"model.layers.17.feed_forward.w2.weight": "model-00008-of-00021.safetensors",
"model.layers.17.feed_forward.w3.weight": "model-00008-of-00021.safetensors",
"model.layers.17.ffn_norm.weight": "model-00008-of-00021.safetensors",
"model.layers.18.attention.wo.weight": "model-00008-of-00021.safetensors",
"model.layers.18.attention.wqkv.weight": "model-00008-of-00021.safetensors",
"model.layers.18.attention_norm.weight": "model-00009-of-00021.safetensors",
"model.layers.18.feed_forward.w1.weight": "model-00008-of-00021.safetensors",
"model.layers.18.feed_forward.w2.weight": "model-00009-of-00021.safetensors",
"model.layers.18.feed_forward.w3.weight": "model-00009-of-00021.safetensors",
"model.layers.18.ffn_norm.weight": "model-00009-of-00021.safetensors",
"model.layers.19.attention.wo.weight": "model-00009-of-00021.safetensors",
"model.layers.19.attention.wqkv.weight": "model-00009-of-00021.safetensors",
"model.layers.19.attention_norm.weight": "model-00009-of-00021.safetensors",
"model.layers.19.feed_forward.w1.weight": "model-00009-of-00021.safetensors",
"model.layers.19.feed_forward.w2.weight": "model-00009-of-00021.safetensors",
"model.layers.19.feed_forward.w3.weight": "model-00009-of-00021.safetensors",
"model.layers.19.ffn_norm.weight": "model-00009-of-00021.safetensors",
"model.layers.2.attention.wo.weight": "model-00002-of-00021.safetensors",
"model.layers.2.attention.wqkv.weight": "model-00002-of-00021.safetensors",
"model.layers.2.attention_norm.weight": "model-00002-of-00021.safetensors",
"model.layers.2.feed_forward.w1.weight": "model-00002-of-00021.safetensors",
"model.layers.2.feed_forward.w2.weight": "model-00002-of-00021.safetensors",
"model.layers.2.feed_forward.w3.weight": "model-00002-of-00021.safetensors",
"model.layers.2.ffn_norm.weight": "model-00002-of-00021.safetensors",
"model.layers.20.attention.wo.weight": "model-00009-of-00021.safetensors",
"model.layers.20.attention.wqkv.weight": "model-00009-of-00021.safetensors",
"model.layers.20.attention_norm.weight": "model-00009-of-00021.safetensors",
"model.layers.20.feed_forward.w1.weight": "model-00009-of-00021.safetensors",
"model.layers.20.feed_forward.w2.weight": "model-00009-of-00021.safetensors",
"model.layers.20.feed_forward.w3.weight": "model-00009-of-00021.safetensors",
"model.layers.20.ffn_norm.weight": "model-00009-of-00021.safetensors",
"model.layers.21.attention.wo.weight": "model-00010-of-00021.safetensors",
"model.layers.21.attention.wqkv.weight": "model-00010-of-00021.safetensors",
"model.layers.21.attention_norm.weight": "model-00010-of-00021.safetensors",
"model.layers.21.feed_forward.w1.weight": "model-00010-of-00021.safetensors",
"model.layers.21.feed_forward.w2.weight": "model-00010-of-00021.safetensors",
"model.layers.21.feed_forward.w3.weight": "model-00010-of-00021.safetensors",
"model.layers.21.ffn_norm.weight": "model-00010-of-00021.safetensors",
"model.layers.22.attention.wo.weight": "model-00010-of-00021.safetensors",
"model.layers.22.attention.wqkv.weight": "model-00010-of-00021.safetensors",
"model.layers.22.attention_norm.weight": "model-00010-of-00021.safetensors",
"model.layers.22.feed_forward.w1.weight": "model-00010-of-00021.safetensors",
"model.layers.22.feed_forward.w2.weight": "model-00010-of-00021.safetensors",
"model.layers.22.feed_forward.w3.weight": "model-00010-of-00021.safetensors",
"model.layers.22.ffn_norm.weight": "model-00010-of-00021.safetensors",
"model.layers.23.attention.wo.weight": "model-00010-of-00021.safetensors",
"model.layers.23.attention.wqkv.weight": "model-00010-of-00021.safetensors",
"model.layers.23.attention_norm.weight": "model-00011-of-00021.safetensors",
"model.layers.23.feed_forward.w1.weight": "model-00010-of-00021.safetensors",
"model.layers.23.feed_forward.w2.weight": "model-00011-of-00021.safetensors",
"model.layers.23.feed_forward.w3.weight": "model-00011-of-00021.safetensors",
"model.layers.23.ffn_norm.weight": "model-00011-of-00021.safetensors",
"model.layers.24.attention.wo.weight": "model-00011-of-00021.safetensors",
"model.layers.24.attention.wqkv.weight": "model-00011-of-00021.safetensors",
"model.layers.24.attention_norm.weight": "model-00011-of-00021.safetensors",
"model.layers.24.feed_forward.w1.weight": "model-00011-of-00021.safetensors",
"model.layers.24.feed_forward.w2.weight": "model-00011-of-00021.safetensors",
"model.layers.24.feed_forward.w3.weight": "model-00011-of-00021.safetensors",
"model.layers.24.ffn_norm.weight": "model-00011-of-00021.safetensors",
"model.layers.25.attention.wo.weight": "model-00011-of-00021.safetensors",
"model.layers.25.attention.wqkv.weight": "model-00011-of-00021.safetensors",
"model.layers.25.attention_norm.weight": "model-00011-of-00021.safetensors",
"model.layers.25.feed_forward.w1.weight": "model-00011-of-00021.safetensors",
"model.layers.25.feed_forward.w2.weight": "model-00011-of-00021.safetensors",
"model.layers.25.feed_forward.w3.weight": "model-00011-of-00021.safetensors",
"model.layers.25.ffn_norm.weight": "model-00011-of-00021.safetensors",
"model.layers.26.attention.wo.weight": "model-00012-of-00021.safetensors",
"model.layers.26.attention.wqkv.weight": "model-00012-of-00021.safetensors",
"model.layers.26.attention_norm.weight": "model-00012-of-00021.safetensors",
"model.layers.26.feed_forward.w1.weight": "model-00012-of-00021.safetensors",
"model.layers.26.feed_forward.w2.weight": "model-00012-of-00021.safetensors",
"model.layers.26.feed_forward.w3.weight": "model-00012-of-00021.safetensors",
"model.layers.26.ffn_norm.weight": "model-00012-of-00021.safetensors",
"model.layers.27.attention.wo.weight": "model-00012-of-00021.safetensors",
"model.layers.27.attention.wqkv.weight": "model-00012-of-00021.safetensors",
"model.layers.27.attention_norm.weight": "model-00012-of-00021.safetensors",
"model.layers.27.feed_forward.w1.weight": "model-00012-of-00021.safetensors",
"model.layers.27.feed_forward.w2.weight": "model-00012-of-00021.safetensors",
"model.layers.27.feed_forward.w3.weight": "model-00012-of-00021.safetensors",
"model.layers.27.ffn_norm.weight": "model-00012-of-00021.safetensors",
"model.layers.28.attention.wo.weight": "model-00012-of-00021.safetensors",
"model.layers.28.attention.wqkv.weight": "model-00012-of-00021.safetensors",
"model.layers.28.attention_norm.weight": "model-00013-of-00021.safetensors",
"model.layers.28.feed_forward.w1.weight": "model-00012-of-00021.safetensors",
"model.layers.28.feed_forward.w2.weight": "model-00013-of-00021.safetensors",
"model.layers.28.feed_forward.w3.weight": "model-00013-of-00021.safetensors",
"model.layers.28.ffn_norm.weight": "model-00013-of-00021.safetensors",
"model.layers.29.attention.wo.weight": "model-00013-of-00021.safetensors",
"model.layers.29.attention.wqkv.weight": "model-00013-of-00021.safetensors",
"model.layers.29.attention_norm.weight": "model-00013-of-00021.safetensors",
"model.layers.29.feed_forward.w1.weight": "model-00013-of-00021.safetensors",
"model.layers.29.feed_forward.w2.weight": "model-00013-of-00021.safetensors",
"model.layers.29.feed_forward.w3.weight": "model-00013-of-00021.safetensors",
"model.layers.29.ffn_norm.weight": "model-00013-of-00021.safetensors",
"model.layers.3.attention.wo.weight": "model-00002-of-00021.safetensors",
"model.layers.3.attention.wqkv.weight": "model-00002-of-00021.safetensors",
"model.layers.3.attention_norm.weight": "model-00003-of-00021.safetensors",
"model.layers.3.feed_forward.w1.weight": "model-00002-of-00021.safetensors",
"model.layers.3.feed_forward.w2.weight": "model-00003-of-00021.safetensors",
"model.layers.3.feed_forward.w3.weight": "model-00003-of-00021.safetensors",
"model.layers.3.ffn_norm.weight": "model-00003-of-00021.safetensors",
"model.layers.30.attention.wo.weight": "model-00013-of-00021.safetensors",
"model.layers.30.attention.wqkv.weight": "model-00013-of-00021.safetensors",
"model.layers.30.attention_norm.weight": "model-00013-of-00021.safetensors",
"model.layers.30.feed_forward.w1.weight": "model-00013-of-00021.safetensors",
"model.layers.30.feed_forward.w2.weight": "model-00013-of-00021.safetensors",
"model.layers.30.feed_forward.w3.weight": "model-00013-of-00021.safetensors",
"model.layers.30.ffn_norm.weight": "model-00013-of-00021.safetensors",
"model.layers.31.attention.wo.weight": "model-00014-of-00021.safetensors",
"model.layers.31.attention.wqkv.weight": "model-00014-of-00021.safetensors",
"model.layers.31.attention_norm.weight": "model-00014-of-00021.safetensors",
"model.layers.31.feed_forward.w1.weight": "model-00014-of-00021.safetensors",
"model.layers.31.feed_forward.w2.weight": "model-00014-of-00021.safetensors",
"model.layers.31.feed_forward.w3.weight": "model-00014-of-00021.safetensors",
"model.layers.31.ffn_norm.weight": "model-00014-of-00021.safetensors",
"model.layers.32.attention.wo.weight": "model-00014-of-00021.safetensors",
"model.layers.32.attention.wqkv.weight": "model-00014-of-00021.safetensors",
"model.layers.32.attention_norm.weight": "model-00014-of-00021.safetensors",
"model.layers.32.feed_forward.w1.weight": "model-00014-of-00021.safetensors",
"model.layers.32.feed_forward.w2.weight": "model-00014-of-00021.safetensors",
"model.layers.32.feed_forward.w3.weight": "model-00014-of-00021.safetensors",
"model.layers.32.ffn_norm.weight": "model-00014-of-00021.safetensors",
"model.layers.33.attention.wo.weight": "model-00014-of-00021.safetensors",
"model.layers.33.attention.wqkv.weight": "model-00014-of-00021.safetensors",
"model.layers.33.attention_norm.weight": "model-00015-of-00021.safetensors",
"model.layers.33.feed_forward.w1.weight": "model-00014-of-00021.safetensors",
"model.layers.33.feed_forward.w2.weight": "model-00015-of-00021.safetensors",
"model.layers.33.feed_forward.w3.weight": "model-00015-of-00021.safetensors",
"model.layers.33.ffn_norm.weight": "model-00015-of-00021.safetensors",
"model.layers.34.attention.wo.weight": "model-00015-of-00021.safetensors",
"model.layers.34.attention.wqkv.weight": "model-00015-of-00021.safetensors",
"model.layers.34.attention_norm.weight": "model-00015-of-00021.safetensors",
"model.layers.34.feed_forward.w1.weight": "model-00015-of-00021.safetensors",
"model.layers.34.feed_forward.w2.weight": "model-00015-of-00021.safetensors",
"model.layers.34.feed_forward.w3.weight": "model-00015-of-00021.safetensors",
"model.layers.34.ffn_norm.weight": "model-00015-of-00021.safetensors",
"model.layers.35.attention.wo.weight": "model-00015-of-00021.safetensors",
"model.layers.35.attention.wqkv.weight": "model-00015-of-00021.safetensors",
"model.layers.35.attention_norm.weight": "model-00015-of-00021.safetensors",
"model.layers.35.feed_forward.w1.weight": "model-00015-of-00021.safetensors",
"model.layers.35.feed_forward.w2.weight": "model-00015-of-00021.safetensors",
"model.layers.35.feed_forward.w3.weight": "model-00015-of-00021.safetensors",
"model.layers.35.ffn_norm.weight": "model-00015-of-00021.safetensors",
"model.layers.36.attention.wo.weight": "model-00016-of-00021.safetensors",
"model.layers.36.attention.wqkv.weight": "model-00016-of-00021.safetensors",
"model.layers.36.attention_norm.weight": "model-00016-of-00021.safetensors",
"model.layers.36.feed_forward.w1.weight": "model-00016-of-00021.safetensors",
"model.layers.36.feed_forward.w2.weight": "model-00016-of-00021.safetensors",
"model.layers.36.feed_forward.w3.weight": "model-00016-of-00021.safetensors",
"model.layers.36.ffn_norm.weight": "model-00016-of-00021.safetensors",
"model.layers.37.attention.wo.weight": "model-00016-of-00021.safetensors",
"model.layers.37.attention.wqkv.weight": "model-00016-of-00021.safetensors",
"model.layers.37.attention_norm.weight": "model-00016-of-00021.safetensors",
"model.layers.37.feed_forward.w1.weight": "model-00016-of-00021.safetensors",
"model.layers.37.feed_forward.w2.weight": "model-00016-of-00021.safetensors",
"model.layers.37.feed_forward.w3.weight": "model-00016-of-00021.safetensors",
"model.layers.37.ffn_norm.weight": "model-00016-of-00021.safetensors",
"model.layers.38.attention.wo.weight": "model-00016-of-00021.safetensors",
"model.layers.38.attention.wqkv.weight": "model-00016-of-00021.safetensors",
"model.layers.38.attention_norm.weight": "model-00017-of-00021.safetensors",
"model.layers.38.feed_forward.w1.weight": "model-00016-of-00021.safetensors",
"model.layers.38.feed_forward.w2.weight": "model-00017-of-00021.safetensors",
"model.layers.38.feed_forward.w3.weight": "model-00017-of-00021.safetensors",
"model.layers.38.ffn_norm.weight": "model-00017-of-00021.safetensors",
"model.layers.39.attention.wo.weight": "model-00017-of-00021.safetensors",
"model.layers.39.attention.wqkv.weight": "model-00017-of-00021.safetensors",
"model.layers.39.attention_norm.weight": "model-00017-of-00021.safetensors",
"model.layers.39.feed_forward.w1.weight": "model-00017-of-00021.safetensors",
"model.layers.39.feed_forward.w2.weight": "model-00017-of-00021.safetensors",
"model.layers.39.feed_forward.w3.weight": "model-00017-of-00021.safetensors",
"model.layers.39.ffn_norm.weight": "model-00017-of-00021.safetensors",
"model.layers.4.attention.wo.weight": "model-00003-of-00021.safetensors",
"model.layers.4.attention.wqkv.weight": "model-00003-of-00021.safetensors",
"model.layers.4.attention_norm.weight": "model-00003-of-00021.safetensors",
"model.layers.4.feed_forward.w1.weight": "model-00003-of-00021.safetensors",
"model.layers.4.feed_forward.w2.weight": "model-00003-of-00021.safetensors",
"model.layers.4.feed_forward.w3.weight": "model-00003-of-00021.safetensors",
"model.layers.4.ffn_norm.weight": "model-00003-of-00021.safetensors",
"model.layers.40.attention.wo.weight": "model-00017-of-00021.safetensors",
"model.layers.40.attention.wqkv.weight": "model-00017-of-00021.safetensors",
"model.layers.40.attention_norm.weight": "model-00017-of-00021.safetensors",
"model.layers.40.feed_forward.w1.weight": "model-00017-of-00021.safetensors",
"model.layers.40.feed_forward.w2.weight": "model-00017-of-00021.safetensors",
"model.layers.40.feed_forward.w3.weight": "model-00017-of-00021.safetensors",
"model.layers.40.ffn_norm.weight": "model-00017-of-00021.safetensors",
"model.layers.41.attention.wo.weight": "model-00018-of-00021.safetensors",
"model.layers.41.attention.wqkv.weight": "model-00018-of-00021.safetensors",
"model.layers.41.attention_norm.weight": "model-00018-of-00021.safetensors",
"model.layers.41.feed_forward.w1.weight": "model-00018-of-00021.safetensors",
"model.layers.41.feed_forward.w2.weight": "model-00018-of-00021.safetensors",
"model.layers.41.feed_forward.w3.weight": "model-00018-of-00021.safetensors",
"model.layers.41.ffn_norm.weight": "model-00018-of-00021.safetensors",
"model.layers.42.attention.wo.weight": "model-00018-of-00021.safetensors",
"model.layers.42.attention.wqkv.weight": "model-00018-of-00021.safetensors",
"model.layers.42.attention_norm.weight": "model-00018-of-00021.safetensors",
"model.layers.42.feed_forward.w1.weight": "model-00018-of-00021.safetensors",
"model.layers.42.feed_forward.w2.weight": "model-00018-of-00021.safetensors",
"model.layers.42.feed_forward.w3.weight": "model-00018-of-00021.safetensors",
"model.layers.42.ffn_norm.weight": "model-00018-of-00021.safetensors",
"model.layers.43.attention.wo.weight": "model-00018-of-00021.safetensors",
"model.layers.43.attention.wqkv.weight": "model-00018-of-00021.safetensors",
"model.layers.43.attention_norm.weight": "model-00019-of-00021.safetensors",
"model.layers.43.feed_forward.w1.weight": "model-00018-of-00021.safetensors",
"model.layers.43.feed_forward.w2.weight": "model-00019-of-00021.safetensors",
"model.layers.43.feed_forward.w3.weight": "model-00019-of-00021.safetensors",
"model.layers.43.ffn_norm.weight": "model-00019-of-00021.safetensors",
"model.layers.44.attention.wo.weight": "model-00019-of-00021.safetensors",
"model.layers.44.attention.wqkv.weight": "model-00019-of-00021.safetensors",
"model.layers.44.attention_norm.weight": "model-00019-of-00021.safetensors",
"model.layers.44.feed_forward.w1.weight": "model-00019-of-00021.safetensors",
"model.layers.44.feed_forward.w2.weight": "model-00019-of-00021.safetensors",
"model.layers.44.feed_forward.w3.weight": "model-00019-of-00021.safetensors",
"model.layers.44.ffn_norm.weight": "model-00019-of-00021.safetensors",
"model.layers.45.attention.wo.weight": "model-00019-of-00021.safetensors",
"model.layers.45.attention.wqkv.weight": "model-00019-of-00021.safetensors",
"model.layers.45.attention_norm.weight": "model-00019-of-00021.safetensors",
"model.layers.45.feed_forward.w1.weight": "model-00019-of-00021.safetensors",
"model.layers.45.feed_forward.w2.weight": "model-00019-of-00021.safetensors",
"model.layers.45.feed_forward.w3.weight": "model-00019-of-00021.safetensors",
"model.layers.45.ffn_norm.weight": "model-00019-of-00021.safetensors",
"model.layers.46.attention.wo.weight": "model-00020-of-00021.safetensors",
"model.layers.46.attention.wqkv.weight": "model-00020-of-00021.safetensors",
"model.layers.46.attention_norm.weight": "model-00020-of-00021.safetensors",
"model.layers.46.feed_forward.w1.weight": "model-00020-of-00021.safetensors",
"model.layers.46.feed_forward.w2.weight": "model-00020-of-00021.safetensors",
"model.layers.46.feed_forward.w3.weight": "model-00020-of-00021.safetensors",
"model.layers.46.ffn_norm.weight": "model-00020-of-00021.safetensors",
"model.layers.47.attention.wo.weight": "model-00020-of-00021.safetensors",
"model.layers.47.attention.wqkv.weight": "model-00020-of-00021.safetensors",
"model.layers.47.attention_norm.weight": "model-00020-of-00021.safetensors",
"model.layers.47.feed_forward.w1.weight": "model-00020-of-00021.safetensors",
"model.layers.47.feed_forward.w2.weight": "model-00020-of-00021.safetensors",
"model.layers.47.feed_forward.w3.weight": "model-00020-of-00021.safetensors",
"model.layers.47.ffn_norm.weight": "model-00020-of-00021.safetensors",
"model.layers.5.attention.wo.weight": "model-00003-of-00021.safetensors",
"model.layers.5.attention.wqkv.weight": "model-00003-of-00021.safetensors",
"model.layers.5.attention_norm.weight": "model-00003-of-00021.safetensors",
"model.layers.5.feed_forward.w1.weight": "model-00003-of-00021.safetensors",
"model.layers.5.feed_forward.w2.weight": "model-00003-of-00021.safetensors",
"model.layers.5.feed_forward.w3.weight": "model-00003-of-00021.safetensors",
"model.layers.5.ffn_norm.weight": "model-00003-of-00021.safetensors",
"model.layers.6.attention.wo.weight": "model-00004-of-00021.safetensors",
"model.layers.6.attention.wqkv.weight": "model-00004-of-00021.safetensors",
"model.layers.6.attention_norm.weight": "model-00004-of-00021.safetensors",
"model.layers.6.feed_forward.w1.weight": "model-00004-of-00021.safetensors",
"model.layers.6.feed_forward.w2.weight": "model-00004-of-00021.safetensors",
"model.layers.6.feed_forward.w3.weight": "model-00004-of-00021.safetensors",
"model.layers.6.ffn_norm.weight": "model-00004-of-00021.safetensors",
"model.layers.7.attention.wo.weight": "model-00004-of-00021.safetensors",
"model.layers.7.attention.wqkv.weight": "model-00004-of-00021.safetensors",
"model.layers.7.attention_norm.weight": "model-00004-of-00021.safetensors",
"model.layers.7.feed_forward.w1.weight": "model-00004-of-00021.safetensors",
"model.layers.7.feed_forward.w2.weight": "model-00004-of-00021.safetensors",
"model.layers.7.feed_forward.w3.weight": "model-00004-of-00021.safetensors",
"model.layers.7.ffn_norm.weight": "model-00004-of-00021.safetensors",
"model.layers.8.attention.wo.weight": "model-00004-of-00021.safetensors",
"model.layers.8.attention.wqkv.weight": "model-00004-of-00021.safetensors",
"model.layers.8.attention_norm.weight": "model-00005-of-00021.safetensors",
"model.layers.8.feed_forward.w1.weight": "model-00004-of-00021.safetensors",
"model.layers.8.feed_forward.w2.weight": "model-00005-of-00021.safetensors",
"model.layers.8.feed_forward.w3.weight": "model-00005-of-00021.safetensors",
"model.layers.8.ffn_norm.weight": "model-00005-of-00021.safetensors",
"model.layers.9.attention.wo.weight": "model-00005-of-00021.safetensors",
"model.layers.9.attention.wqkv.weight": "model-00005-of-00021.safetensors",
"model.layers.9.attention_norm.weight": "model-00005-of-00021.safetensors",
"model.layers.9.feed_forward.w1.weight": "model-00005-of-00021.safetensors",
"model.layers.9.feed_forward.w2.weight": "model-00005-of-00021.safetensors",
"model.layers.9.feed_forward.w3.weight": "model-00005-of-00021.safetensors",
"model.layers.9.ffn_norm.weight": "model-00005-of-00021.safetensors",
"model.norm.weight": "model-00020-of-00021.safetensors",
"model.tok_embeddings.weight": "model-00001-of-00021.safetensors",
"output.weight": "model-00021-of-00021.safetensors"
}
}

1391
modeling_internlm2.py Normal file
View File

File diff suppressed because it is too large Load Diff

6
special_tokens_map.json Normal file
View File

@@ -0,0 +1,6 @@
{
"bos_token": "<s>",
"eos_token": "</s>",
"pad_token": "</s>",
"unk_token": "<unk>"
}

236
tokenization_internlm2.py Normal file
View File

@@ -0,0 +1,236 @@
# coding=utf-8
# Copyright (c) The InternLM team and The HuggingFace Inc. team. All rights reserved.
#
# This code is based on transformers/src/transformers/models/llama/tokenization_llama.py
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Tokenization classes for InternLM."""
import os
from shutil import copyfile
from typing import Any, Dict, List, Optional, Tuple
import sentencepiece as spm
from transformers.tokenization_utils import PreTrainedTokenizer
from transformers.utils import logging
logger = logging.get_logger(__name__)
VOCAB_FILES_NAMES = {"vocab_file": "./tokenizer.model"}
PRETRAINED_VOCAB_FILES_MAP = {}
# Modified from transformers.model.llama.tokenization_llama.LlamaTokenizer
class InternLM2Tokenizer(PreTrainedTokenizer):
"""
Construct a InternLM2 tokenizer. Based on byte-level Byte-Pair-Encoding.
Args:
vocab_file (`str`):
Path to the vocabulary file.
"""
vocab_files_names = VOCAB_FILES_NAMES
pretrained_vocab_files_map = PRETRAINED_VOCAB_FILES_MAP
model_input_names = ["input_ids", "attention_mask"]
_auto_class = "AutoTokenizer"
def __init__(
self,
vocab_file,
unk_token="<unk>",
bos_token="<s>",
eos_token="</s>",
pad_token="</s>",
sp_model_kwargs: Optional[Dict[str, Any]] = None,
add_bos_token=True,
add_eos_token=False,
decode_with_prefix_space=False,
clean_up_tokenization_spaces=False,
**kwargs,
):
self.sp_model_kwargs = {} if sp_model_kwargs is None else sp_model_kwargs
self.vocab_file = vocab_file
self.add_bos_token = add_bos_token
self.add_eos_token = add_eos_token
self.decode_with_prefix_space = decode_with_prefix_space
self.sp_model = spm.SentencePieceProcessor(**self.sp_model_kwargs)
self.sp_model.Load(vocab_file)
self._no_prefix_space_tokens = None
super().__init__(
bos_token=bos_token,
eos_token=eos_token,
unk_token=unk_token,
pad_token=pad_token,
clean_up_tokenization_spaces=clean_up_tokenization_spaces,
**kwargs,
)
@property
def no_prefix_space_tokens(self):
if self._no_prefix_space_tokens is None:
vocab = self.convert_ids_to_tokens(list(range(self.vocab_size)))
self._no_prefix_space_tokens = {i for i, tok in enumerate(vocab) if not tok.startswith("▁")}
return self._no_prefix_space_tokens
@property
def vocab_size(self):
"""Returns vocab size"""
return self.sp_model.get_piece_size()
@property
def bos_token_id(self) -> Optional[int]:
return self.sp_model.bos_id()
@property
def eos_token_id(self) -> Optional[int]:
return self.sp_model.eos_id()
def get_vocab(self):
"""Returns vocab as a dict"""
vocab = {self.convert_ids_to_tokens(i): i for i in range(self.vocab_size)}
vocab.update(self.added_tokens_encoder)
return vocab
def _tokenize(self, text):
"""Returns a tokenized string."""
return self.sp_model.encode(text, out_type=str)
def _convert_token_to_id(self, token):
"""Converts a token (str) in an id using the vocab."""
return self.sp_model.piece_to_id(token)
def _convert_id_to_token(self, index):
"""Converts an index (integer) in a token (str) using the vocab."""
token = self.sp_model.IdToPiece(index)
return token
def _maybe_add_prefix_space(self, tokens, decoded):
if tokens and tokens[0] not in self.no_prefix_space_tokens:
return " " + decoded
else:
return decoded
def convert_tokens_to_string(self, tokens):
"""Converts a sequence of tokens (string) in a single string."""
current_sub_tokens = []
out_string = ""
prev_is_special = False
for token in tokens:
# make sure that special tokens are not decoded using sentencepiece model
if token in self.all_special_tokens:
if not prev_is_special:
out_string += " "
out_string += self.sp_model.decode(current_sub_tokens) + token
prev_is_special = True
current_sub_tokens = []
else:
current_sub_tokens.append(token)
prev_is_special = False
out_string += self.sp_model.decode(current_sub_tokens)
out_string = self.clean_up_tokenization(out_string)
out_string = self._maybe_add_prefix_space(tokens=tokens, decoded=out_string)
return out_string[1:]
def save_vocabulary(self, save_directory, filename_prefix: Optional[str] = None) -> Tuple[str]:
"""
Save the vocabulary and special tokens file to a directory.
Args:
save_directory (`str`):
The directory in which to save the vocabulary.
Returns:
`Tuple(str)`: Paths to the files saved.
"""
if not os.path.isdir(save_directory):
logger.error(f"Vocabulary path ({save_directory}) should be a directory")
return
out_vocab_file = os.path.join(
save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
)
if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file) and os.path.isfile(self.vocab_file):
copyfile(self.vocab_file, out_vocab_file)
elif not os.path.isfile(self.vocab_file):
with open(out_vocab_file, "wb") as fi:
content_spiece_model = self.sp_model.serialized_model_proto()
fi.write(content_spiece_model)
return (out_vocab_file,)
def build_inputs_with_special_tokens(self, token_ids_0, token_ids_1=None):
if self.add_bos_token:
bos_token_ids = [self.bos_token_id]
else:
bos_token_ids = []
output = bos_token_ids + token_ids_0
if token_ids_1 is not None:
output = output + token_ids_1
if self.add_eos_token:
output = output + [self.eos_token_id]
return output
def get_special_tokens_mask(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None, already_has_special_tokens: bool = False
) -> List[int]:
"""
Retrieve sequence ids from a token list that has no special tokens added. This method is called when adding
special tokens using the tokenizer `prepare_for_model` method.
Args:
token_ids_0 (`List[int]`):
List of IDs.
token_ids_1 (`List[int]`, *optional*):
Optional second list of IDs for sequence pairs.
already_has_special_tokens (`bool`, *optional*, defaults to `False`):
Whether or not the token list is already formatted with special tokens for the model.
Returns:
`List[int]`: A list of integers in the range [0, 1]: 1 for a special token, 0 for a sequence token.
"""
if already_has_special_tokens:
return super().get_special_tokens_mask(
token_ids_0=token_ids_0, token_ids_1=token_ids_1, already_has_special_tokens=True
)
if token_ids_1 is None:
return [1] + ([0] * len(token_ids_0)) + [1]
return [1] + ([0] * len(token_ids_0)) + [1, 1] + ([0] * len(token_ids_1)) + [1]
def create_token_type_ids_from_sequences(
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
) -> List[int]:
"""
Create a mask from the two sequences passed to be used in a sequence-pair classification task. T5 does not make
use of token type ids, therefore a list of zeros is returned.
Args:
token_ids_0 (`List[int]`):
List of IDs.
token_ids_1 (`List[int]`, *optional*):
Optional second list of IDs for sequence pairs.
Returns:
`List[int]`: List of zeros.
"""
eos = [self.eos_token_id]
if token_ids_1 is None:
return len(token_ids_0 + eos) * [0]
return len(token_ids_0 + eos + token_ids_1 + eos) * [0]

214
tokenization_internlm2_fast.py Normal file
View File

@@ -0,0 +1,214 @@
# coding=utf-8
# Copyright (c) The InternLM team and The HuggingFace Inc. team. All rights reserved.
#
# This code is based on transformers/src/transformers/models/llama/tokenization_llama_fast.py
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Tokenization Fast class for InternLM."""
import os
from shutil import copyfile
from typing import Any, Dict, Optional, Tuple
from tokenizers import processors, decoders, Tokenizer, normalizers
from tokenizers.models import BPE
from transformers.tokenization_utils_fast import PreTrainedTokenizerFast
from transformers.utils import logging
from transformers.convert_slow_tokenizer import (
SLOW_TO_FAST_CONVERTERS,
SpmConverter,
SentencePieceExtractor,
)
from .tokenization_internlm2 import InternLM2Tokenizer
logger = logging.get_logger(__name__)
VOCAB_FILES_NAMES = {"vocab_file": "./tokenizer.model"}
# Modified from transformers.convert_slow_tokenizer.LlamaConverter
class InternLM2Converter(SpmConverter):
handle_byte_fallback = True
def vocab(self, proto):
vocab = [
("<unk>", 0.0),
("<s>", 0.0),
("</s>", 0.0),
]
vocab += [(piece.piece, piece.score) for piece in proto.pieces[3:]]
return vocab
def unk_id(self, proto):
unk_id = 0
return unk_id
def decoder(self, replacement, add_prefix_space):
return decoders.Sequence(
[
decoders.Replace("▁", " "),
decoders.ByteFallback(),
decoders.Fuse(),
decoders.Strip(content=" ", left=1),
]
)
def tokenizer(self, proto):
model_type = proto.trainer_spec.model_type
vocab_scores = self.vocab(proto)
# special tokens
added_tokens = self.original_tokenizer.added_tokens_decoder
for i in range(len(vocab_scores)):
piece, score = vocab_scores[i]
if i in added_tokens:
vocab_scores[i] = (added_tokens[i].content, score)
if model_type == 1:
raise RuntimeError("InternLM2 is supposed to be a BPE model!")
elif model_type == 2:
_, merges = SentencePieceExtractor(self.original_tokenizer.vocab_file).extract(vocab_scores)
bpe_vocab = {word: i for i, (word, _score) in enumerate(vocab_scores)}
tokenizer = Tokenizer(
BPE(bpe_vocab, merges, unk_token=proto.trainer_spec.unk_piece, fuse_unk=True, byte_fallback=True)
)
tokenizer.add_special_tokens(
[ added_token for index, added_token in added_tokens.items()]
)
else:
raise Exception(
"You're trying to run a `Unigram` model but you're file was trained with a different algorithm"
)
return tokenizer
def normalizer(self, proto):
normalizers_list = []
if proto.normalizer_spec.add_dummy_prefix:
normalizers_list.append(normalizers.Prepend(prepend="▁"))
normalizers_list.append(normalizers.Replace(pattern=" ", content="▁"))
return normalizers.Sequence(normalizers_list)
def pre_tokenizer(self, replacement, add_prefix_space):
return None
SLOW_TO_FAST_CONVERTERS["InternLM2Tokenizer"] = InternLM2Converter
# Modified from transformers.model.llama.tokenization_llama_fast.LlamaTokenizerFast -> InternLM2TokenizerFast
class InternLM2TokenizerFast(PreTrainedTokenizerFast):
vocab_files_names = VOCAB_FILES_NAMES
slow_tokenizer_class = InternLM2Tokenizer
padding_side = "left"
model_input_names = ["input_ids", "attention_mask"]
_auto_class = "AutoTokenizer"
def __init__(
self,
vocab_file,
unk_token="<unk>",
bos_token="<s>",
eos_token="</s>",
pad_token="</s>",
sp_model_kwargs: Optional[Dict[str, Any]] = None,
add_bos_token=True,
add_eos_token=False,
decode_with_prefix_space=False,
clean_up_tokenization_spaces=False,
**kwargs,
):
super().__init__(
vocab_file=vocab_file,
unk_token=unk_token,
bos_token=bos_token,
eos_token=eos_token,
pad_token=pad_token,
sp_model_kwargs=sp_model_kwargs,
add_bos_token=add_bos_token,
add_eos_token=add_eos_token,
decode_with_prefix_space=decode_with_prefix_space,
clean_up_tokenization_spaces=clean_up_tokenization_spaces,
**kwargs,
)
self._add_bos_token = add_bos_token
self._add_eos_token = add_eos_token
self.update_post_processor()
self.vocab_file = vocab_file
@property
def can_save_slow_tokenizer(self) -> bool:
return os.path.isfile(self.vocab_file) if self.vocab_file else False
def update_post_processor(self):
"""
Updates the underlying post processor with the current `bos_token` and `eos_token`.
"""
bos = self.bos_token
bos_token_id = self.bos_token_id
if bos is None and self.add_bos_token:
raise ValueError("add_bos_token = True but bos_token = None")
eos = self.eos_token
eos_token_id = self.eos_token_id
if eos is None and self.add_eos_token:
raise ValueError("add_eos_token = True but eos_token = None")
single = f"{(bos+':0 ') if self.add_bos_token else ''}$A:0{(' '+eos+':0') if self.add_eos_token else ''}"
pair = f"{single}{(' '+bos+':1') if self.add_bos_token else ''} $B:1{(' '+eos+':1') if self.add_eos_token else ''}"
special_tokens = []
if self.add_bos_token:
special_tokens.append((bos, bos_token_id))
if self.add_eos_token:
special_tokens.append((eos, eos_token_id))
self._tokenizer.post_processor = processors.TemplateProcessing(
single=single, pair=pair, special_tokens=special_tokens
)
@property
def add_eos_token(self):
return self._add_eos_token
@property
def add_bos_token(self):
return self._add_bos_token
@add_eos_token.setter
def add_eos_token(self, value):
self._add_eos_token = value
self.update_post_processor()
@add_bos_token.setter
def add_bos_token(self, value):
self._add_bos_token = value
self.update_post_processor()
def save_vocabulary(self, save_directory: str, filename_prefix: Optional[str] = None) -> Tuple[str]:
if not self.can_save_slow_tokenizer:
raise ValueError(
"Your fast tokenizer does not have the necessary information to save the vocabulary for a slow "
"tokenizer."
)
if not os.path.isdir(save_directory):
logger.error(f"Vocabulary path ({save_directory}) should be a directory")
return
out_vocab_file = os.path.join(
save_directory, (filename_prefix + "-" if filename_prefix else "") + VOCAB_FILES_NAMES["vocab_file"]
)
if os.path.abspath(self.vocab_file) != os.path.abspath(out_vocab_file):
copyfile(self.vocab_file, out_vocab_file)
return (out_vocab_file,)

3
tokenizer.model Normal file
View File

@@ -0,0 +1,3 @@
version https://git-lfs.github.com/spec/v1
oid sha256:f868398fc4e05ee1e8aeba95ddf18ddcc45b8bce55d5093bead5bbf80429b48b
size 1477754

90
tokenizer_config.json Normal file
View File

@@ -0,0 +1,90 @@
{
"auto_map": {
"AutoTokenizer": [
"tokenization_internlm2.InternLM2Tokenizer",
"tokenization_internlm2_fast.InternLM2TokenizerFast"
]
},
"bos_token": "<s>",
"clean_up_tokenization_spaces": false,
"eos_token": "</s>",
"model_max_length": 1000000000000000019884624838656,
"pad_token": "</s>",
"tokenizer_class": "InternLM2Tokenizer",
"unk_token": "<unk>",
"added_tokens_decoder": {
"0": {
"content": "<unk>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"1": {
"content": "<s>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"2": {
"content": "</s>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"92543": {
"content": "<|im_start|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"92542": {
"content": "<|im_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"92541": {
"content": "<|action_start|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"92540": {
"content": "<|action_end|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"92539": {
"content": "<|interpreter|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
},
"92538": {
"content": "<|plugin|>",
"lstrip": false,
"normalized": false,
"rstrip": false,
"single_word": false,
"special": true
}
},
"chat_template": "{{ bos_token }}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}"
}