File size: 2,693 Bytes
ebe31fb 879a7e8 cd07f68 ebe31fb 879a7e8 cd07f68 879a7e8 cd07f68 879a7e8 cd07f68 879a7e8 f880b12 879a7e8 cd07f68 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 | ---
library_name: transformers
license: apache-2.0
base_model:
- lkevinzc/Llama-3.2-3B-NuminaQA
---
# Llama-3.2-3B-Oat-Zero
## Links
- 📜 [Paper](https://github.com/sail-sg/understand-r1-zero/blob/main/understand-r1-zero.pdf)
- 💻 [GitHub](https://github.com/sail-sg/understand-r1-zero)
- 🤗 [Oat-Zero Collection](https://huggingface.co/collections/sail/oat-zero-understanding-r1-zero-like-training-67dcdb07b9f3eb05f1501c4a)
## Introduction
This model is trained by the minimalist R1-Zero recipe introduced in our paper:
- **Algorithm**: Dr. DRPO
- **Data**: level 3-5 questions from MATH dataset
- **Base model**: [lkevinzc/Llama-3.2-3B-NuminaQA](https://huggingface.co/lkevinzc/Llama-3.2-3B-NuminaQA)
- **Template**: R1
Evaluation results on widely used math benchmarks are shown below:
<img src="https://raw.githubusercontent.com/sail-sg/understand-r1-zero/refs/heads/main/assets/benchmark_table.png" width=100%/>
## Usage
```python
import vllm
def apply_qwen_math_template(question: str):
return (
"<|im_start|>system\nPlease reason step by step, and put your final answer within \\boxed{}.<|im_end|>\n<|im_start|>user\n"
+ question
+ "<|im_end|>\n<|im_start|>assistant\n"
)
def apply_r1_template(question: str):
return (
"A conversation between User and Assistant. The User asks a question, and the Assistant solves it. The Assistant first thinks about the reasoning process in the mind and then provides the User with the answer. "
"The reasoning process is enclosed within <think> </think> and answer is enclosed within <answer> </answer> tags, respectively, i.e., <think> reasoning process here </think> <answer> answer here </answer>.\nUser: "
+ question
+ "\nAssistant: <think>"
)
model_name = "sail/Llama-3.2-3B-Oat-Zero"
sampling_params = vllm.SamplingParams(
n=1,
temperature=0,
top_p=1,
max_tokens=3000,
)
model = vllm.LLM(
model_name,
max_model_len=4096,
dtype="bfloat16",
enable_prefix_caching=True,
)
if "Llama-3.2-3B-Oat-Zero" in model_name:
apply_template = apply_r1_template
else:
apply_template = apply_qwen_math_template
prompts = [
"How many positive whole-number divisors does 196 have?"
]
prompts = list(map(apply_template, prompts))
outputs = model.generate(prompts, sampling_params)
print(outputs)
```
## Citation
```latex
@article{liu2025understanding,
title={Understanding r1-zero-like training: A critical perspective},
author={Liu, Zichen and Chen, Changyu and Li, Wenjun and Qi, Penghui and Pang, Tianyu and Du, Chao and Lee, Wee Sun and Lin, Min},
journal={arXiv preprint arXiv:2503.20783},
year={2025}
}
```
|