metadata

license: apache-2.0
datasets:
  - 6cf/liveideabench
language:
  - en
base_model:
  - Qwen/QwQ-32B-Preview
tags:
  - chemistry
  - biology
  - climate
  - medical

IdeaWhiz Model Card 🧠

Model Summary 🔬

IdeaWhiz is a fine-tuned version of QwQ-32B-Preview, specifically optimized for scientific creativity and step-by-step reasoning. The model leverages the LiveIdeaBench dataset to enhance its capabilities in generating novel scientific ideas and hypotheses.

Key Features 🌟

Base Model: QwQ-32B-Preview 🔋
Training Dataset: LiveIdeaBench 📊
Main Focus: Scientific creativity and idea generation 💡
Reasoning Style: o1-style step-by-step reasoning ⚡

Intended Use 🎯

Scientific hypothesis generation 🧪
Creative problem-solving in research 🔍
Step-by-step scientific reasoning 📝
Research direction brainstorming 🌱

Model Performance Compared to QwQ-32B-Preview

Quickstart 🚀

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "6cf/QwQ-32B-Preview-IdeaWhiz-v1"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)



prompt = """I'll be submitting your next responses to a "Good Scientific Idea" expert review panel. If they consider your idea to be a good one, you'll receive a reward. Your assigned keyword is: "cancer". You may provide background information. The idea MUST be within 100 words (including background information). (Note: good scientific ideas should be novel, verifiable, practically valuable, and able to advance the field.). NOTE: You MUST give your answer after **Final Idea:**
"""


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=4096
)
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]

Example Outputs

Alright, so I have this task to come up with a good scientific idea related to cancer,
and it has to be novel, verifiable, practically valuable, and able to advance the field.
Plus, it needs to be within 100 words, including any background information.
That's a pretty tight constraint, but I think I can manage. First, ...

...

**Final Idea:** propose a novel approach to cancer treatment by developing personalized
cancer vaccines tailored to each patient's unique tumor mutations. By integrating machine
learning algorithms with comprehensive omics data, including genomics, epigenetics,
transcriptomics, and proteomics, this method aims to identify and prioritize the most
immunogenic tumor-specific antigens. This integrated strategy could enhance vaccine
efficacy and represents an advancement in precision medicine for cancer treatment.

Traning Dataset

🤖💡 LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context

Dataset

Paper

If you use this model, please cite:

@article{ruan2024liveideabench,
title={LiveIdeaBench: Evaluating LLMs' Scientific Creativity and Idea Generation with Minimal Context},
author={Ruan, Kai and Wang, Xuan and Hong, Jixiang and Sun, Hao},
journal={arXiv preprint arXiv:2412.17596},
year={2024}
}