初始化项目，由ModelHub XC社区提供模型
Model: SantiagoC/palindrome-grpo Source: Original Platform
This commit is contained in:
ModelHub XC
36
.gitattributes
vendored
Normal file
36
.gitattributes
vendored
Normal file
@@ -0,0 +1,36 @@
|
||||
*.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
|
||||
tokenizer.json filter=lfs diff=lfs merge=lfs -text
|
||||
90
README.md
Normal file
90
README.md
Normal file
@@ -0,0 +1,90 @@
|
||||
---
|
||||
base_model: Qwen/Qwen2.5-0.5B-Instruct
|
||||
library_name: transformers
|
||||
model_name: palindrome-grpo
|
||||
tags:
|
||||
- generated_from_trainer
|
||||
- hf_jobs
|
||||
- trl
|
||||
- trackio:https://SantiagoC-mlintern-palindrm.hf.space?project=palindrome-grpo&runs=palindrome-grpo-v1&sidebar=collapsed
|
||||
- grpo
|
||||
- ml-intern
|
||||
licence: license
|
||||
---
|
||||
|
||||
# Model Card for palindrome-grpo
|
||||
|
||||
This model is a fine-tuned version of [Qwen/Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct).
|
||||
It has been trained using [TRL](https://github.com/huggingface/trl).
|
||||
|
||||
## Quick start
|
||||
|
||||
```python
|
||||
from transformers import pipeline
|
||||
|
||||
question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
|
||||
generator = pipeline("text-generation", model="SantiagoC/palindrome-grpo", device="cuda")
|
||||
output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
|
||||
print(output["generated_text"])
|
||||
```
|
||||
|
||||
## Training procedure
|
||||
|
||||
|
||||
[<img src="https://raw.githubusercontent.com/gradio-app/trackio/refs/heads/main/trackio/assets/badge.png" alt="Visualize in Trackio" title="Visualize in Trackio" width="150" height="24"/>](https://SantiagoC-mlintern-palindrm.hf.space?project=palindrome-grpo&runs=palindrome-grpo-v1&sidebar=collapsed)
|
||||
|
||||
|
||||
This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).
|
||||
|
||||
### Framework versions
|
||||
|
||||
- TRL: 1.3.0
|
||||
- Transformers: 5.8.0
|
||||
- Pytorch: 2.11.0
|
||||
- Datasets: 4.8.5
|
||||
- Tokenizers: 0.22.2
|
||||
|
||||
## Citations
|
||||
|
||||
Cite GRPO as:
|
||||
|
||||
```bibtex
|
||||
@article{shao2024deepseekmath,
|
||||
title = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
|
||||
author = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
|
||||
year = 2024,
|
||||
eprint = {arXiv:2402.03300},
|
||||
}
|
||||
```
|
||||
|
||||
Cite TRL as:
|
||||
|
||||
```bibtex
|
||||
@software{vonwerra2020trl,
|
||||
title = {{TRL: Transformers Reinforcement Learning}},
|
||||
author = {von Werra, Leandro and Belkada, Younes and Tunstall, Lewis and Beeching, Edward and Thrush, Tristan and Lambert, Nathan and Huang, Shengyi and Rasul, Kashif and Gallouédec, Quentin},
|
||||
license = {Apache-2.0},
|
||||
url = {https://github.com/huggingface/trl},
|
||||
year = {2020}
|
||||
}
|
||||
```
|
||||
|
||||
<!-- ml-intern-provenance -->
|
||||
## Generated by ML Intern
|
||||
|
||||
This model repository was generated by [ML Intern](https://github.com/huggingface/ml-intern), an agent for machine learning research and development on the Hugging Face Hub.
|
||||
|
||||
- Try ML Intern: https://smolagents-ml-intern.hf.space
|
||||
- Source code: https://github.com/huggingface/ml-intern
|
||||
|
||||
## Usage
|
||||
|
||||
```python
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
|
||||
model_id = 'SantiagoC/palindrome-grpo'
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
||||
model = AutoModelForCausalLM.from_pretrained(model_id)
|
||||
```
|
||||
|
||||
For non-causal architectures, replace `AutoModelForCausalLM` with the appropriate `AutoModel` class.
|
||||
54
chat_template.jinja
Normal file
54
chat_template.jinja
Normal file
@@ -0,0 +1,54 @@
|
||||
{%- if tools %}
|
||||
{{- '<|im_start|>system\n' }}
|
||||
{%- if messages[0]['role'] == 'system' %}
|
||||
{{- messages[0]['content'] }}
|
||||
{%- else %}
|
||||
{{- 'You are Qwen, created by Alibaba Cloud. You are a helpful assistant.' }}
|
||||
{%- endif %}
|
||||
{{- "\n\n# 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' }}
|
||||
{%- else %}
|
||||
{{- '<|im_start|>system\nYou are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>\n' }}
|
||||
{%- endif %}
|
||||
{%- endif %}
|
||||
{%- for message in messages %}
|
||||
{%- if (message.role == "user") or (message.role == "system" and not loop.first) or (message.role == "assistant" and not message.tool_calls) %}
|
||||
{{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
|
||||
{%- elif message.role == "assistant" %}
|
||||
{{- '<|im_start|>' + message.role }}
|
||||
{%- if message.content %}
|
||||
{{- '\n' + message.content }}
|
||||
{%- endif %}
|
||||
{%- for tool_call in message.tool_calls %}
|
||||
{%- if tool_call.function is defined %}
|
||||
{%- set tool_call = tool_call.function %}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_call>\n{"name": "' }}
|
||||
{{- tool_call.name }}
|
||||
{{- '", "arguments": ' }}
|
||||
{{- tool_call.arguments | tojson }}
|
||||
{{- '}\n</tool_call>' }}
|
||||
{%- endfor %}
|
||||
{{- '<|im_end|>\n' }}
|
||||
{%- elif message.role == "tool" %}
|
||||
{%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != "tool") %}
|
||||
{{- '<|im_start|>user' }}
|
||||
{%- endif %}
|
||||
{{- '\n<tool_response>\n' }}
|
||||
{{- message.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' }}
|
||||
{%- endif %}
|
||||
57
config.json
Normal file
57
config.json
Normal file
@@ -0,0 +1,57 @@
|
||||
{
|
||||
"architectures": [
|
||||
"Qwen2ForCausalLM"
|
||||
],
|
||||
"attention_dropout": 0.0,
|
||||
"bos_token_id": null,
|
||||
"dtype": "float32",
|
||||
"eos_token_id": 151645,
|
||||
"hidden_act": "silu",
|
||||
"hidden_size": 896,
|
||||
"initializer_range": 0.02,
|
||||
"intermediate_size": 4864,
|
||||
"layer_types": [
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention",
|
||||
"full_attention"
|
||||
],
|
||||
"max_position_embeddings": 32768,
|
||||
"max_window_layers": 21,
|
||||
"model_type": "qwen2",
|
||||
"num_attention_heads": 14,
|
||||
"num_hidden_layers": 24,
|
||||
"num_key_value_heads": 2,
|
||||
"pad_token_id": 151643,
|
||||
"rms_norm_eps": 1e-06,
|
||||
"rope_parameters": {
|
||||
"rope_theta": 1000000.0,
|
||||
"rope_type": "default"
|
||||
},
|
||||
"sliding_window": null,
|
||||
"tie_word_embeddings": true,
|
||||
"transformers_version": "5.8.0",
|
||||
"use_cache": false,
|
||||
"use_sliding_window": false,
|
||||
"vocab_size": 151936
|
||||
}
|
||||
13
generation_config.json
Normal file
13
generation_config.json
Normal file
@@ -0,0 +1,13 @@
|
||||
{
|
||||
"do_sample": true,
|
||||
"eos_token_id": [
|
||||
151645,
|
||||
151643
|
||||
],
|
||||
"pad_token_id": 151643,
|
||||
"repetition_penalty": 1.1,
|
||||
"temperature": 0.7,
|
||||
"top_k": 20,
|
||||
"top_p": 0.8,
|
||||
"transformers_version": "5.8.0"
|
||||
}
|
||||
3
model.safetensors
Normal file
3
model.safetensors
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:abe93aaae15122a73eef519e5c0809deca51d59a88efd7b10fa41350ce134daa
|
||||
size 1976163472
|
||||
3
tokenizer.json
Normal file
3
tokenizer.json
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:3fd169731d2cbde95e10bf356d66d5997fd885dd8dbb6fb4684da3f23b2585d8
|
||||
size 11421892
|
||||
32
tokenizer_config.json
Normal file
32
tokenizer_config.json
Normal file
@@ -0,0 +1,32 @@
|
||||
{
|
||||
"add_prefix_space": false,
|
||||
"backend": "tokenizers",
|
||||
"bos_token": null,
|
||||
"clean_up_tokenization_spaces": false,
|
||||
"eos_token": "<|im_end|>",
|
||||
"errors": "replace",
|
||||
"extra_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|>"
|
||||
],
|
||||
"is_local": false,
|
||||
"local_files_only": false,
|
||||
"model_max_length": 131072,
|
||||
"pad_token": "<|endoftext|>",
|
||||
"padding_side": "left",
|
||||
"split_special_tokens": false,
|
||||
"tokenizer_class": "Qwen2Tokenizer",
|
||||
"truncation_side": "left",
|
||||
"unk_token": null
|
||||
}
|
||||
272
train_grpo.py
Normal file
272
train_grpo.py
Normal file
@@ -0,0 +1,272 @@
|
||||
"""
|
||||
GRPO training script for palindrome generation.
|
||||
|
||||
Trains a small LLM (Qwen2.5-0.5B-Instruct) to generate palindromes
|
||||
given a theme using Group Relative Policy Optimization.
|
||||
|
||||
Reward function:
|
||||
1. Palindrome accuracy (continuous partial-credit: proportion of matching
|
||||
mirrored character pairs)
|
||||
2. Length bonus (longer palindromes = more reward)
|
||||
3. Theme relevance (keyword overlap between theme and generated text)
|
||||
|
||||
reward = α * palindrome_accuracy + β * length_bonus + γ * theme_relevance
|
||||
"""
|
||||
import re
|
||||
import os
|
||||
from datasets import load_dataset, Dataset
|
||||
from trl import GRPOConfig, GRPOTrainer
|
||||
from transformers import TrainerCallback
|
||||
import trackio
|
||||
|
||||
|
||||
# ── Reward function ─────────────────────────────────────────────────────
|
||||
|
||||
def _normalize(text: str) -> str:
|
||||
"""Strip non-alphanumeric and lowercase."""
|
||||
return re.sub(r'[^a-zA-Z0-9]', '', text).lower()
|
||||
|
||||
|
||||
def _palindrome_accuracy(text: str) -> float:
|
||||
"""
|
||||
Continuous palindrome score: proportion of matching mirrored pairs.
|
||||
|
||||
Perfect palindrome → 1.0
|
||||
"ractar" (close to racecar) → partial credit
|
||||
Empty / single char → 0.0 or 1.0 respectively
|
||||
"""
|
||||
cleaned = _normalize(text)
|
||||
n = len(cleaned)
|
||||
if n == 0:
|
||||
return 0.0
|
||||
if n == 1:
|
||||
return 1.0
|
||||
|
||||
matches = 0
|
||||
pairs = n // 2
|
||||
for i in range(pairs):
|
||||
if cleaned[i] == cleaned[n - 1 - i]:
|
||||
matches += 1
|
||||
return matches / pairs
|
||||
|
||||
|
||||
def _length_bonus(text: str, max_len: int = 80) -> float:
|
||||
"""
|
||||
Normalized length reward: longer palindromes score higher.
|
||||
Capped at max_len characters.
|
||||
"""
|
||||
cleaned = _normalize(text)
|
||||
return min(len(cleaned), max_len) / max_len
|
||||
|
||||
|
||||
def _theme_relevance(text: str, theme: str) -> float:
|
||||
"""
|
||||
Simple keyword overlap: does the palindrome contain the theme word
|
||||
or substrings of it? Returns 0.0 - 1.0.
|
||||
"""
|
||||
text_lower = _normalize(text)
|
||||
theme_clean = _normalize(theme)
|
||||
|
||||
if len(theme_clean) == 0:
|
||||
return 0.0
|
||||
|
||||
# Exact match of the full theme word
|
||||
if theme_clean in text_lower:
|
||||
return 1.0
|
||||
|
||||
# Partial match: theme word split into n-grams
|
||||
if len(theme_clean) > 3:
|
||||
bigrams = set(theme_clean[i:i + 2] for i in range(len(theme_clean) - 1))
|
||||
match_count = sum(1 for bg in bigrams if bg in text_lower)
|
||||
return match_count / len(bigrams)
|
||||
|
||||
return 0.0
|
||||
|
||||
|
||||
def palindrome_reward(
|
||||
prompts,
|
||||
completions,
|
||||
completion_ids,
|
||||
theme=None,
|
||||
**kwargs,
|
||||
) -> list[float]:
|
||||
"""
|
||||
GRPO reward function for palindrome generation.
|
||||
|
||||
Args:
|
||||
prompts: list[list[dict]] — conversational prompts
|
||||
completions: list[list[dict]] — generated completions
|
||||
completion_ids: list[list[int]] — token IDs
|
||||
theme: list[str] — theme column from dataset (forwarded via **kwargs)
|
||||
|
||||
Returns:
|
||||
list[float] — one reward per sample
|
||||
"""
|
||||
# Reward weights (tunable)
|
||||
alpha = 0.6 # palindrome accuracy weight
|
||||
beta = 0.25 # length bonus weight
|
||||
gamma = 0.15 # theme relevance weight
|
||||
|
||||
rewards = []
|
||||
for i, completion in enumerate(completions):
|
||||
# Extract text from conversational format
|
||||
if isinstance(completion, list):
|
||||
text = completion[0]["content"].strip()
|
||||
else:
|
||||
text = str(completion).strip()
|
||||
|
||||
acc = _palindrome_accuracy(text)
|
||||
length = _length_bonus(text)
|
||||
|
||||
# Theme from dataset column
|
||||
theme_text = ""
|
||||
if theme is not None:
|
||||
# theme list is flattened: [t*n_generations for t in batch_themes]
|
||||
theme_text = theme[i] if i < len(theme) else ""
|
||||
|
||||
theme_rel = _theme_relevance(text, theme_text) if theme_text else 0.0
|
||||
|
||||
reward = alpha * acc + beta * length + gamma * theme_rel
|
||||
rewards.append(reward)
|
||||
|
||||
return rewards
|
||||
|
||||
|
||||
# ── Trackio alert callback ──────────────────────────────────────────────
|
||||
|
||||
class PalindromeAlertCallback(TrainerCallback):
|
||||
"""Log Trackio alerts on key training events."""
|
||||
|
||||
def on_train_begin(self, args, state, control, **kwargs):
|
||||
trackio.alert(
|
||||
"Training started",
|
||||
f"Model: Qwen/Qwen2.5-0.5B-Instruct | "
|
||||
f"num_generations={args.num_generations} | "
|
||||
f"temperature={args.temperature} | "
|
||||
f"lr={args.learning_rate} | "
|
||||
f"batch_size={args.per_device_train_batch_size}×"
|
||||
f"{args.gradient_accumulation_steps}",
|
||||
level="INFO",
|
||||
)
|
||||
|
||||
def on_train_end(self, args, state, control, **kwargs):
|
||||
trackio.alert(
|
||||
"Training complete",
|
||||
f"Model pushed to {args.hub_model_id}",
|
||||
level="INFO",
|
||||
)
|
||||
|
||||
def on_log(self, args, state, control, logs=None, **kwargs):
|
||||
if logs is None:
|
||||
return
|
||||
|
||||
# Reward monitoring
|
||||
reward = logs.get("reward")
|
||||
if reward is not None:
|
||||
if reward < 0.1 and state.global_step > 50:
|
||||
trackio.alert(
|
||||
"Low reward",
|
||||
f"reward={reward:.4f} at step {state.global_step} — "
|
||||
"model not producing palindromes yet. Consider increasing "
|
||||
"temperature for more exploration.",
|
||||
level="WARN",
|
||||
)
|
||||
elif 0.1 <= reward < 0.5 and state.global_step > 50:
|
||||
trackio.alert(
|
||||
"Moderate reward",
|
||||
f"reward={reward:.4f} at step {state.global_step} — "
|
||||
"model starting to learn. Continue training.",
|
||||
level="INFO",
|
||||
)
|
||||
elif reward >= 0.7:
|
||||
trackio.alert(
|
||||
"High reward milestone",
|
||||
f"reward={reward:.4f} at step {state.global_step} — "
|
||||
"model producing good palindromes!",
|
||||
level="INFO",
|
||||
)
|
||||
|
||||
# KL divergence monitoring
|
||||
kl = logs.get("kl")
|
||||
if kl is not None and kl > 0.5:
|
||||
trackio.alert(
|
||||
"High KL divergence",
|
||||
f"kl={kl:.4f} at step {state.global_step} — "
|
||||
"policy diverging too far. Consider lowering learning rate.",
|
||||
level="WARN",
|
||||
)
|
||||
|
||||
# Loss monitoring
|
||||
loss = logs.get("loss")
|
||||
if loss is not None:
|
||||
if loss > 10:
|
||||
trackio.alert(
|
||||
"High loss",
|
||||
f"loss={loss:.4f} at step {state.global_step} — "
|
||||
"loss spike detected. Check for instability, "
|
||||
"consider reducing learning rate by 0.5×.",
|
||||
level="WARN",
|
||||
)
|
||||
|
||||
|
||||
# ── Main ────────────────────────────────────────────────────────────────
|
||||
|
||||
def main():
|
||||
# Load dataset
|
||||
dataset_path = os.getenv("DATASET_PATH", "SantiagoC/palindrome-prompts")
|
||||
dataset = load_dataset(dataset_path, split="train")
|
||||
|
||||
print(f"Dataset size: {len(dataset)} samples")
|
||||
print(f"Columns: {dataset.column_names}")
|
||||
print(f"First prompt: {dataset[0]['prompt']}")
|
||||
|
||||
# Training config
|
||||
training_args = GRPOConfig(
|
||||
output_dir="./palindrome-grpo-output",
|
||||
# ── GRPO specific ──
|
||||
num_generations=4,
|
||||
max_completion_length=128,
|
||||
temperature=0.9,
|
||||
beta=0.0, # No reference model KL penalty
|
||||
scale_rewards=False, # Dr.GRPO recommendation
|
||||
loss_type="dapo",
|
||||
mask_truncated_completions=True,
|
||||
# ── Standard training ──
|
||||
learning_rate=1e-6,
|
||||
num_train_epochs=6,
|
||||
per_device_train_batch_size=2,
|
||||
gradient_accumulation_steps=4,
|
||||
bf16=True,
|
||||
gradient_checkpointing=True,
|
||||
# ── Logging ──
|
||||
logging_steps=10,
|
||||
logging_first_step=True,
|
||||
save_steps=100,
|
||||
save_total_limit=3,
|
||||
disable_tqdm=True,
|
||||
report_to="trackio",
|
||||
# ── Hub push ──
|
||||
push_to_hub=True,
|
||||
hub_model_id=os.getenv("HUB_MODEL_ID", "SantiagoC/palindrome-grpo"),
|
||||
run_name=os.getenv("RUN_NAME", "palindrome-grpo-v1"),
|
||||
project="palindrome-grpo",
|
||||
trackio_space_id=os.getenv("TRACKIO_SPACE_ID", "SantiagoC/mlintern-palindrm"),
|
||||
)
|
||||
|
||||
trainer = GRPOTrainer(
|
||||
model="Qwen/Qwen2.5-0.5B-Instruct",
|
||||
reward_funcs=palindrome_reward,
|
||||
args=training_args,
|
||||
train_dataset=dataset,
|
||||
callbacks=[PalindromeAlertCallback()],
|
||||
)
|
||||
|
||||
trainer.train()
|
||||
|
||||
# Final push
|
||||
trainer.save_model()
|
||||
trainer.push_to_hub()
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
3
training_args.bin
Normal file
3
training_args.bin
Normal file
@@ -0,0 +1,3 @@
|
||||
version https://git-lfs.github.com/spec/v1
|
||||
oid sha256:c9b6b4e5d0e7fab9407ca337c0d19353e218e56a19a3d70b81e0333423d618ca
|
||||
size 7313
|
||||
Reference in New Issue
Block a user