Intel/Llama-3.3-70B-Instruct-AutoRound-Recipe

Model Details

This model card is for mxfp8/mxfp4/mxfp4_mixed quantization of meta-llama/Llama-3.3-70B-Instruct based on intel/auto-round. The models are not able to be published due to the license limitation. Please follow the INC example README to generate and evaluate the low precision models.

How to Use

The step-by-step README of quantization and evaluation can be found in Intel Neural Compressor Examples.

Evaluate Results

Task	backend	BF16	MXFP8	MXFP4(mixed)
hellaswag	vllm	0.6661	0.6660	0.6567
piqa	vllm	0.8357	0.8324	0.8303
mmlu_llama	vllm	0.8336	0.8326	0.8291
gsm8k_llama(strict)	vllm	0.9492	0.9439	0.9416
average	vllm	0.8211	0.8187	0.8144

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:

• Intel Neural Compressor
• AutoRound

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.