Intel Neural Compressor Release 3.8

• Highlights
• Features
• Improvements
• Validated Hardware
• Validated Configurations

Highlights

• Introduced new JAX framework experimental support

Features

• Support FP8 quantization for Keras/JAX (experimental)
• Support FP8 KV cache static quantization (experimental)
• Support FP8 Attention static quantization (experimental)

Improvements

• New Gemma3 FP8 PTQ example for Keras/JAX
• New ViT FP8 PTQ example for Keras/JAX
• Llama 3 series MXFP4 / MXFP8 PTQ example with FP8 KV & Attention
• Llama 4 Scout MXFP4 / MXFP8 PTQ example with FP8 KV & Attention
• Qwen3 MXFP4 / MXFP8 PTQ example with FP8 KV & Attention
• DeepSeek R1 MXFP4 / MXFP8 PTQ example with FP8 KV & Attention
• Transformers v5 support
• Removal of deprecated 2.x API

Validated Hardware 

• Intel Gaudi Al Accelerators (Gaudi 2 and 3)
• Intel Xeon Scalable processor (4th, 5th and 6th Gen)
• Intel® Arc™ B-Series Graphics GPU (B580 and B60)

Validated Configurations

• Ubuntu 24.04 & Win 11
• Python 3.11, 3.12, 3.13
• PyTorch 2.9, 2.10
• JAX 0.9

Notes

• It is recommended to use version v3.8 or later to mitigate code CVEs.