Support weight-only quantization with quantized operators in intel-extension-for-transformers.#455

Merged

echarlaix merged 69 commits intohuggingface:mainfrom PenghuiCheng:penghuic/weight_only_with_itrex

Mar 27, 2024

+331-191

Commits on Jan 16, 2024

Commits on Jan 17, 2024

Update example
PenghuiCheng
committed

Commits on Feb 21, 2024

merge from main branch
PenghuiCheng
committed

Commits on Mar 13, 2024

Update optimum/intel/neural_compressor/quantization.py

PenghuiCheng
and
echarlaix
authored
[OV]: Fixed inference after 4 bit weight compression (huggingface#569 )

authored and
PenghuiCheng
committed
Updated docs with load_in_4bit (huggingface#558 )

authored and
PenghuiCheng
committed
Update Transformers dependency requirements (huggingface#571 )

echarlaix
authored and
PenghuiCheng
committed
Fix compatibility for latest transformers release (huggingface#570 )

echarlaix
authored and
PenghuiCheng
committed
Deprecate compression options (huggingface#565 )

authored and
PenghuiCheng
committed
Add default quantization int4 config for Mixtral-8x7B (huggingface#576 )

ljaljushkin
authored and
PenghuiCheng
committed
Update stable diffusion example requirements (huggingface#579 )

helena-intel
authored and
PenghuiCheng
committed
Fix collecting duplicate tensors in quantization calibration dataset (huggingface#577 )

nikita-savelyevv
authored and
PenghuiCheng
committed
Save an openvino config summarizing all information related to quantization when saving model (huggingface#578 )

echarlaix
authored and
PenghuiCheng
committed
Fix warning (huggingface#582 )

echarlaix
authored and
PenghuiCheng
committed
Add reference to the temporary directory for windows fix (huggingface#581 )

echarlaix
authored and
PenghuiCheng
committed
Fix documentation (huggingface#583 )

echarlaix
authored and
PenghuiCheng
committed
Add llama test model to cover MQA (huggingface#585 )

jiqing-feng
authored and
PenghuiCheng
committed
Include nncf in openvino extra (huggingface#586 )

eaidova
authored and
PenghuiCheng
committed
Fix title documentation (huggingface#588 )

echarlaix
authored and
PenghuiCheng
committed
Update OpenVINO documentation links in README.md (huggingface#587 )

kblaszczak-intel
authored and
PenghuiCheng
committed
Fix default int8 quantization for CLI (huggingface#592 )

echarlaix
authored and
PenghuiCheng
committed
Change model output parameter to last_hidden_states for IPEXModel (huggingface#589 )

jiqing-feng
authored and
PenghuiCheng
committed
Add IPEX model patcher (huggingface#567 )

authored and
PenghuiCheng
committed
Updates weight quantization section in the docs (huggingface#593 )

AlexKoff88
authored and
PenghuiCheng
committed
Remove accelerate and onnxruntime from required dependencies (huggingface#590 )

echarlaix
authored and
PenghuiCheng
committed
Fix OpenVINO image classification examples (huggingface#598 )

echarlaix
authored and
PenghuiCheng
committed
Fix weights compression for OPenVINO models (huggingface#596 )

eaidova
authored and
PenghuiCheng
committed
Fix default ov config (huggingface#600 )

echarlaix
authored and
PenghuiCheng
committed
Add warning for transformers>=4.38 and OpenVINO 2024.0 (huggingface#599 )

helena-intel
authored and
PenghuiCheng
committed
Add hybrid quantization for StableDiffusion pipelines (huggingface#584 )

l-bat
authored and
PenghuiCheng
committed
Show device name in _print_compiled_model_properties (huggingface#541 )

authored and
PenghuiCheng
committed
Update code with comments
PenghuiCheng
committed
Fixed pylint error
PenghuiCheng
committed
Merge from main branch
PenghuiCheng
committed
Update optimum/intel/neural_compressor/configuration.py

PenghuiCheng
and
echarlaix
authored
Fixed example and UT for weight-only quantization
PenghuiCheng
committed
Fixed pre-ci test error
PenghuiCheng
committed
Fixed pre-ci test error
PenghuiCheng
committed

Commits on Mar 17, 2024

Commits on Mar 18, 2024

Fixed UT error
PenghuiCheng
committed

Commits on Mar 23, 2024

Commits on Mar 24, 2024

Commits on Mar 25, 2024

Commits on Mar 26, 2024