/Qwen-WisdomVast

Qwen-WisdomVast is a large model trained on 1 million high-quality Chinese multi-turn SFT data, 200,000 English multi-turn SFT data, and 2,000 single-turn self-cognition data, using the training methods of DORA and LORA+ based on Qwen1.5-7B as the base. Compared to Qwen1.5-7B-Chat, it has improved mathematical abilities by 5.16%, 12.8% on the Human

Primary LanguagePythonApache License 2.0Apache-2.0

中文  ｜  English



Qwen-WisdomVast (千问-智瀚)


GitHub Contributors

Introduce

Qwen-WisdomVast is a large model trained on 1 million high-quality Chinese multi-turn SFT data, 200,000 English multi-turn SFT data, and 2,000 single-turn self-cognition data, using the training methods of DORA and LORA+ based on Qwen1.5-7B as the base. Compared to Qwen1.5-7B-Chat, it has improved mathematical abilities by 5.16%, 12.8% on the HumanEval dataset, 11.6% on the MBPP dataset, and 12.44% on the BBH dataset. The performance on all evaluations is shown in the table below.

DEMO

Evaluation Results

Model MMLU C-Eval GSM8K MATH HumanEval MBPP BBH
Qwen1.5-7B-Chat 60.88 70.18 54.13 7.96 31.10 15.00 31.67
Qwen-WisdomVast 57.09 70.82 51.93 13.12 43.90 26.60 44.11

Explanation:

Since the official evaluation performance of Qwen1.5-7B-Chat has not been disclosed, we conducted our own testing using opencompass and obtained the above results.

Qwen-WisdomVast was tested using the same parameters as Qwen1.5-7B-Chat.

Download Model

Model Download
Qwen1.5-7B 🤗 HuggingFace 🤖 ModelScope
Qwen-WisdomVast-Lora 🤗 HuggingFace 🤖 ModelScope
Qwen-WisdomVast (Merged Model) 🤗 HuggingFace 🤖 ModelScope

Merge LORA Model (Skippable)

1、Download Qwen1.5-7B

git clone https://www.modelscope.cn/qwen/Qwen1.5-7B.git

2、Download Qwen-WisdomVast-Lora

From ModelScope

git lfs install
git clone https://www.modelscope.cn/seanzhang/Qwen-WisdomVast-Lora.git

From HuggingFace

git lfs install
git clone https://huggingface.co/zhichen/Qwen-WisdomVast-Lora

3、Merge Model

python merge_lora.py \
    --base_model path/to/qwen/Qwen1.5-7B \
    --lora_model path/to/lora/Qwen-WisdomVast-Lora \
    --output_dir ./Qwen-WisdomVast

Download Qwen-WisdomVast (Merged Model)

From ModelScope

git lfs install
git clone https://www.modelscope.cn/seanzhang/Qwen-WisdomVast.git

From HuggingFace

git lfs install
git clone https://huggingface.co/zhichen/Qwen-WisdomVast

CLI DEMO

python cli_demo.py  --model_path ./Qwen-WisdomVast(Replace it with your own merged model path)

WEB DEMO

python web_demo.py  --model_path ./Qwen-WisdomVast(Replace it with your own merged model path)

VLLM WEB DEMO

1、Use vllm deploy model

python -m vllm.entrypoints.openai.api_server --served-model-name Qwen-WisdomVast --model ./Qwen-WisdomVast(Replace it with your own merged model path)

2、This command is executed on the CLI

python vllm_web_demo.py --model Qwen-WisdomVast

Repeat the evaluation results

1、Use vllm deploy openai api server

deploy command:

python -m vllm.entrypoints.openai.api_server --served-model-name Qwen-WisdomVast --model ./Qwen-WisdomVast(Replace it with your own merged model path)

2、Use opencompass framework to eval

Reference: Verify model effects using opencompass

After modifying as described above, copy the eval_qwen_wisdomvast.py file in the opencompass/configs folder

3、Execute test script

python run.py configs/eval_qwen_wisdomvast.py  -w outputs/Qwen-WisdomVast

LICENSE

This project can only be used for research purposes, and the project developer shall not bear any harm or loss caused by the use of this project (including but not limited to data, models, codes, etc.). For details, please refer to DISCLAIMER。

The License agreement of the Qwen-WisdomVast project code is the Apache License 2.0. The code is free for commercial use, and the model weights and data can only be used for research purposes. Please attach a link to Qwen-WisdomVast and the licensing agreement in the product description.

Citation

If you used Qwen-WisdomVast in your research, cite it in the following format:

@misc{Qwen-WisdomVast,
  title={Qwen-WisdomVast},
  author={Zhichen Zhang, Weihan Huang},
  year={2024},
  howpublished={\url{https://github.com/seanzhang-zhichen/Qwen-WisdomVast}},
}

Acknowledgement

QwenLM/Qwen1.5
hiyouga/LLaMA-Factory
shibing624/MedicalGPT
modelscope/swift

Star History

Star History Chart