neural-compressor
neural-compressor copied to clipboard
SOTA low-bit LLM quantization (INT8/FP8/INT4/FP4/NF4) & sparsity; leading model compression techniques on TensorFlow, PyTorch, and ONNX Runtime
[Line 474 in weight_only.py](https://github.com/intel/neural-compressor/blob/24419c9044fe227ea806db370c1a30272d026f8a/neural_compressor/adaptor/ox_utils/weight_only.py#L474C37-L474C43) is trying to pad the the tensor so that the first dimension is divisible by the block size. The problem is that instead of passing the...
Hi, I ran the resnet based quantization notebook. I am getting the following error. Please suggest me to solve this issue. Used !pip install tensorflow==2.12.0 !pip install neural-compressor==2.3.2 ``` 2024-04-03...
## Type of Change Use google coding style for ONC ## Description detail description ## Expected Behavior & Potential Risk the expected behavior that triggered by this PR ## How...
## Type of Change example API changed or not: no ## Description Update 3.x torch example and enhance 3.x common logger information. Smooth quant uses quantize(), others use prepare() +...
## Type of Change feature or bug fix or documentation or validation or others API changed or not ## Description detail description ## Expected Behavior & Potential Risk the expected...
## Type of Change Smoothquant refactor for 3.x API API changed ## Description - [x] refactor new API - prepare/convert - [ ] add ut for new API, remove unnecessary...
## Type of Change bug fix ## Description https://github.com/intel/neural-compressor/issues/1699 ## Expected Behavior & Potential Risk the expected behavior that triggered by this PR ## How has this PR been tested?...
## Type of Change feature ## Description Support save/load API for WOQ remove export_compressed_model from config ## Expected Behavior & Potential Risk the expected behavior that triggered by this PR...
## Type of Change bug fix ## Description the q_config of tuning history is wrong in performance only mode ## How has this PR been tested? UT
## Type of Change feature ## Description detail description ## Expected Behavior & Potential Risk the expected behavior that triggered by this PR ## How has this PR been tested?...