WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct (RLEIF)
News
[12/19/2023] 🔥 We released WizardMath-7B-V1.1 trained from Mistral-7B, the SOTA 7B math LLM, achieves 83.2 pass@1 on GSM8k, and 33.0 pass@1 on MATH. Use this [Demo] to chat with it.
[12/19/2023] 🔥 WizardMath-7B-V1.1 outperforms ChatGPT 3.5, Gemini Pro, Mixtral MOE, and Claude Instant on GSM8K pass@1.
[12/19/2023] 🔥 WizardMath-7B-V1.1 is comparable with ChatGPT 3.5, Gemini Pro, and surpasses Mixtral MOE on MATH pass@1.
| Model | Checkpoint | Paper | GSM8k | MATH | Demo |
|---|---|---|---|---|---|
| WizardMath-7B-V1.1 | 🤗 HF Link | 📃 [WizardMath] | 83.2 | 33.0 | [Demo] |
| WizardMath-70B-V1.0 | 🤗 HF Link | 📃 [WizardMath] | 81.6 | 22.7 | |
| WizardMath-13B-V1.0 | 🤗 HF Link | 📃 [WizardMath] | 63.9 | 14.0 | |
| WizardMath-7B-V1.0 | 🤗 HF Link | 📃 [WizardMath] | 54.9 | 10.7 |
[12/19/2023] Comparing WizardMath-7B-V1.1 with other open source 7B size math LLMs.
| Model | GSM8k Pass@1 | MATH Pass@1 |
|---|---|---|
| MPT-7B | 6.8 | 3.0 |
| Llama 1-7B | 11.0 | 2.9 |
| Llama 2-7B | 12.3 | 2.8 |
| Yi-6b | 32.6 | 5.8 |
| Mistral-7B | 37.8 | 9.1 |
| Qwen-7b | 47.8 | 9.3 |
| RFT-7B | 50.3 | -- |
| MAmmoTH-7B (COT) | 50.5 | 10.4 |
| WizardMath-7B-V1.0 | 54.9 | 10.7 |
| Abel-7B-001 | 59.7 | 13 |
| MetaMath-7B | 66.5 | 19.8 |
| Arithmo-Mistral-7B | 74.7 | 25.3 |
| MetaMath-Mistral-7B | 77.7 | 28.2 |
| Abel-7B-002 | 80.4 | 29.5 |
| WizardMath-7B-V1.1 | 83.2 | 33.0 |
[12/19/2023] Comparing WizardMath-7B-V1.1 with large open source (30B~70B) LLMs.
| Model | GSM8k Pass@1 | MATH Pass@1 |
|---|---|---|
| Llemma-34B | 51.5 | 25.0 |
| Minerva-62B | 52.4 | 27.6 |
| Llama 2-70B | 56.8 | 13.5 |
| DeepSeek 67B | 63.4 | -- |
| Gork 33B | 62.9 | 23.9 |
| MAmmoTH-70B | 72.4 | 21.1 |
| Yi-34B | 67.9 | 15.9 |
| Mixtral 8x7B | 74.4 | 28.4 |
| MetaMath-70B | 82.3 | 26.6 |
| WizardMath-7B-V1.1 | 83.2 | 33.0 |
❗ Data Contamination Check:
Before model training, we carefully and rigorously checked all the training data, and used multiple deduplication methods to verify and prevent data leakage on GSM8k and MATH test set.
🔥 ❗Note for model system prompts usage:
Please use the same systems prompts strictly with us, and we do not guarantee the accuracy of the quantified versions.
Default version:
"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response:"
CoT Version: （❗For the simple math questions, we do NOT recommend to use the CoT prompt.）
"Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n### Instruction:\n{instruction}\n\n### Response: Let's think step by step."
Inference WizardMath Demo Script
We provide the WizardMath inference demo code here.
Citation
Please cite the repo if you use the data, method or code in this repo.
@article{luo2023wizardmath,
title={WizardMath: Empowering Mathematical Reasoning for Large Language Models via Reinforced Evol-Instruct},
author={Luo, Haipeng and Sun, Qingfeng and Xu, Can and Zhao, Pu and Lou, Jianguang and Tao, Chongyang and Geng, Xiubo and Lin, Qingwei and Chen, Shifeng and Zhang, Dongmei},
journal={arXiv preprint arXiv:2308.09583},
year={2023}
}