README.md

---
license: apache-2.0
---

# QwQ-Math-7B-Persona

## Introduction

QwQ-Math-7B-Persona is finetuned from Qwen2.5-Math-7B-Instruct on 1 million math persona data (see [this paper](https://arxiv.org/abs/2406.20094) for details about how to construct the data).

Currently QwQ-Math-7B-Persona is meant to serve as a draft model for losslessly accelerating the inference of QwQ-32B, but you may also use it as a standalone model.

## Quickstart

Here is a code snippet for using QwQ-Math-7B-Persona to accelerate the inference of QwQ 32B:

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "Qwen/QwQ-32B-Preview",
    torch_dtype="auto",
    device_map={'': 0}
)

draft_model = AutoModelForCausalLM.from_pretrained(
    "Geralt-Targaryen/QwQ-Math-7B-Persona",
    torch_dtype="auto",
    device_map={'': 0}
)

tokenizer = AutoTokenizer.from_pretrained("Qwen/QwQ-32B-Preview")

prompt = "How many r in strawberry."
messages = [
    {"role": "system", "content": "You are a helpful and harmless assistant. You are Qwen developed by Alibaba. You should think step-by-step."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512,
    assistant_model=draft_model
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
```

For the more advanced SVIP draft length policy, please refer to [this GitHub repo](https://github.com/Geralt-Targaryen/SVIP).

## Citation

If you find QwQ-Math-1.5B-Persona to be helpful, please cite the following paper.

```
@misc{zhang2024svip,
      title={Draft Model Knows When to Stop: A Self-Verification Length Policy for Speculative Decoding},
      author={Ziyin Zhang and Jiahao Xu and Tian Liang and Xingyu Chen and Zhiwei He and Rui Wang and Zhaopeng Tu},
      year={2024},
      eprint={2411.18462},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2411.18462},
}
```