- Highlights
- Features
- Improvements
- Validated Hardware
- Validated Configurations
Highlights
- Introduced new low precision data type quantization experimental support, including MXFP8 and MXFP4
Features
- Support MXFP8 Post-Training Quantization (PTQ) on LLM models (experimental)
- Support MXFP8 PTQ on diffusion models (experimental)
- Support MXFP4 PTQ on LLM models (experimental)
- Support Quantization-Aware Training (QAT) on LLM models (experimental)
Improvements
- New LLM example (Llama 3 series) for MXFP4 / MXFP8 PTQ
- New VLM example (Llama 4 Scout) for MXFP4 PTQ
- New diffusion example (Flux) for MXFP8 PTQ
- New LLM example (Llama 3) for MXFP8 QAT
- Static safe check for evaluation function in 2.x API
Validated Hardware
- Intel Gaudi Al Accelerators (Gaudi 2 and 3)
- Intel Xeon Scalable processor (4th, 5th, 6th Gen)
- Intel Core Ultra Processors (Series 1 and 2)
- Intel Data Center GPU Max Series (1550)
- Intel® Arc™ B-Series Graphics GPU (B580)
Validated Configurations
- Ubuntu 24.04 & Win 11
- Python 3.10, 3.11, 3.12
- PyTorch/IPEX 2.7, 2.8