ScaleQuest
Collection
We introduce ScaleQuest, a scalable and novel data synthesis method. Project Page: https://scalequest.github.io/
•
8 items
•
Updated
•
5
We introduce ScaleQuest, a scalable and novel data synthesis method that utilizes small-size open-source models to generate questions from scratch without the need for seed data with complex augmentation constraints.
Math Dataset: link
We release two question generator models and four problem-solving models.
Model | Type | MATH | Olympiad Bench | 🤗 HuggingFace Download Link |
---|---|---|---|---|
ScaleQuest-DeepSeekMath-7B-QGen | question generator | - | - | link |
ScaleQuest-Qwen2-Math-7B-QGen | question generator | - | - | link |
Mistral-7B-ScaleQuest | problem solver | 62.9 | 26.8 | link |
Llama3-8B-ScaleQuest | problem solver | 64.4 | 25.3 | link |
DeepSeekMath-7B-ScaleQuest | problem solver | 66.6 | 29.9 | link |
Qwen2-Math-7B-ScaleQuest | problem solver | 73.4 | 38.5 | link |
Below is an example using ScaleQuest-DeepSeekMath-7B-QGen
from vllm import LLM, SamplingParams
model_name = "dyyyyyyyy/ScaleQuest-DeepSeekMath-7B-QGen"
pre_query_template = "<|begin▁of▁sentence|>User: "
stop_tokens = ["<|begin▁of▁sentence|>", "<|end▁of▁sentence|>"]
llm = LLM(
model=model_name,
tokenizer=model_name,
tensor_parallel_size=1,
max_model_len=4096,
enable_prefix_caching=True,
trust_remote_code=True,
swap_space=16,
gpu_memory_utilization=0.95,
)
sampling_params = SamplingParams(
n=4,
max_tokens=1024,
temperature=1.0,
top_p=0.99,
stop=stop_tokens,
)
outputs = llm.generate(pre_query_template, sampling_params)
# Print the outputs.
for output in outputs:
prompt = output.prompt
for idx, generated_output in enumerate(output.outputs):
generated_text = generated_output.text
print(f"Sample {idx + 1}:")
print(f"Prompt: {prompt!r}")
print(f"Generated text: {generated_text!r}")
print("-" * 50)
@article{ding2024unleashing,
title={Unleashing Reasoning Capability of LLMs via Scalable Question Synthesis from Scratch},
author={Ding, Yuyang and Shi, Xinyu and Liang, Xiaobo and Li, Juntao and Zhu, Qiaoming and Zhang, Min},
journal={https://arxiv.org/abs/2410.18693},
year={2024}
}