INCModel2/MiniMax-M2.7-MXFP4-Mixed-CT-AutoRound
Model Details
This model is a MXFP4 mixed model of MiniMaxAI/MiniMax-M2.7 generated by intel/auto-round with RTN mode. Please follow the license of the original model. | Accuracy (repeats=3) | aime25 | gpqa_diamond | gsm8k | piqa | |----------------------|--------|--------------|--------|--------| | Raw | 0.8556 | 0.8872 | 0.9636 | 0.9400 | | Intel | 0.8889 | 0.8821 | 0.9641 | 0.9436 | | Ratio | 1.0390 | 0.9943 | 1.0005 | 1.0039 |
vllm Infernece Example
~~~bash CUDAVISIBLEDEVICES=3,4,5,7 vllm serve ~/models/minimax-m2.7-mxfp \ --trust-remote-code \ --tensor-parallel-size 4 \ --tool-call-parser minimaxm2 \ --enable-auto-tool-choice \ --reasoning-parser minimaxm2 \ --served-model-name mxfp \ --max-model-len 102400 \ --max-num-seqs 1024 \ --max-num-batched-tokens 32768 \ --enable-chunked-prefill \ --port 8001 ~~~
~~~bash
Prompt generation
curl http://localhost:8000/v1/chat/completions -H "Content-Type: application/json" -d ' { "model": "mxfp", "messages": [ {"role": "system", "content": "You are a helpful assistant."}, {"role": "user", "content": "Write code to fine-tune an LLM."} ], "temperature": 1, "max_tokens": 2048 } '
Accuracy evaluation
evalscope eval --model mxfp --eval-type openaiapi --api-key EMPTY --timeout 36000 --datasets gpqadiamond aime25 gsm8k piqa \ --generation-config '{"temperature":1.0, "topp":0.95, "n":1, "extrabody": { "chattemplatekwargs": { "enablethinking": true, "reasoningeffort": "max"}},"max_tokens":64000}' \ --eval-batch-size 512 --api-url http://127.0.0.1:8001/v1 ~~~
Generate the Model
RTN version
~~~bash auto-round MiniMaxAI/MiniMax-M2.7 --modelfree --scheme MXFP8 --layerconfig {blocksparsemoe:{scheme:MXFP4}} --outputdir minimax-m2.7-mxfp --format llmcompressor ~~~
Ethical Considerations and Limitations
The model can produce factually incorrect output, and should not be relied on to produce factually accurate information. Because of the limitations of the pretrained model and the finetuning datasets, it is possible that this model could generate lewd, biased or otherwise offensive outputs.
Therefore, before deploying any applications of the model, developers should perform safety testing.
Caveats and Recommendations
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model.
Here are a couple of useful links to learn more about Intel's AI software:
Disclaimer
The license on this model does not constitute legal advice. We are not responsible for the actions of third parties who use this model. Please consult an attorney before using this model for commercial purposes.
Cite
@article{cheng2023optimize, title={Optimize weight rounding via signed gradient descent for the quantization of llms}, author={Cheng, Wenhua and Zhang, Weiwei and Shen, Haihao and Cai, Yiyang and He, Xin and Lv, Kaokao and Liu, Yi}, journal={arXiv preprint arXiv:2309.05516}, year={2023} }