improve load_empty_model function. #2079
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
xin3he
approved these changes
Dec 4, 2024
thuang6
approved these changes
Dec 4, 2024
|
Please trigger some extension tests as well. Can ask @XuehaoSun for help. |
|
Memory is fine, but from_pretrained prints some warning.
|
Kaihui-intel
approved these changes
Dec 4, 2024
3 tasks
yiliu30
added a commit
that referenced
this pull request
Feb 14, 2025
* modify 3.x ipex example structure (#1858)
* modify 3.x ipex example structure
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
* add json path
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
* fix for sq
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
* minor fix
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
* Update run_clm_no_trainer.py
* Update run_clm_no_trainer.py
* Update run_clm_no_trainer.py
* minor fix
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
* remove old files
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
* fix act_algo
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
---------
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
Co-authored-by: xinhe <xin3.he@intel.com>
* Improve UT Branch Coverage for TF 3x (#1867)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* [3x] add recommendation examples (#1844)
Signed-off-by: xin3he <xin3.he@intel.com>
* Add PT2E cv&llm example (#1853)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Update SQ/WOQ status (#1869)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
Co-authored-by: chen, suyue <suyue.chen@intel.com>
* Modify WOQ examples structure (#1866)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Signed-off-by: chensuyue <suyue.chen@intel.com>
* update v2.6 release readme (#1871)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Limit numpy versions (#1874)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* fix layer match (#1873)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
* Enhance autotune to return the best `q_model` directly (#1875)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Add op statistics dump for woq (#1876)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* rm cov (#1878)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Add `set_local` support for static quant with pt2e (#1870)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Update the Gaudi container example in the README (#1885)
* support quant_lm_head arg in all WOQ configs (#1881)
Signed-off-by: xin3he <xin3.he@intel.com>
* Fix sql injection for Neural Solution gRPC (#1879)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Remove Gelu Fusion for TF Newapi (#1886)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* Refine HQQ UTs (#1888)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* tmp fix nas deps issue (#1896)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* support auto_host2device on RTN and GPTQ(#1894)
Signed-off-by: He, Xin3 <xin3.he@intel.com>
* remove import pdb (#1897)
Signed-off-by: changwangss <chang1.wang@intel.com>
* Port auto-detect absorb layers for TEQ (#1895)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Remove 1x API (#1865)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Co-authored-by: chen, suyue <suyue.chen@intel.com>
* remove neural insight CI (#1903)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* fix bf16 symbolic_trace bug (#1892)
Description: fix bf16 symbolic_trace bug,
- cause abnormal recursive calling.
- missing necessary attributes
- By moving BF16 fallback ahead of quantization and removing bf16_symbolic_trace, we fix it.
---------
Signed-off-by: xin3he <xin3.he@intel.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
* update fp4_e2m1 mapping list (#1906)
* update fp4_e2m1 mapping list
* Update utility.py
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
---------
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Add docstring for `common` module (#1905)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* support habana fp8 UT test in CI (#1909)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* bump version into 3.0 (#1908)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* implement `incbench` command for ease-of-use benchmark (#1884)
# Description
implement incbench command as entrypoint for ease-of-use benchmark
automatically check numa/socket info and dump it with table for ease-of-understand
supports both Linux and Windows platform
add benchmark documents
dump benchmark summary
add benchmark UTs
# General Use Cases
incbench main.py: run 1 instance on NUMA:0.
incbench --num_i 2 main.py: run 2 instances on NUMA:0.
incbench --num_c 2 main.py: run multi-instances with 2 cores per instance on NUMA:0.
incbench -C 24-47 main.py: run 1 instance on COREs:24-47.
incbench -C 24-47 --num_c 4 main.py: run multi-instances with 4 COREs per instance on COREs:24-47.
---------
Signed-off-by: xin3he <xin3.he@intel.com>
Co-authored-by: chen, suyue <suyue.chen@intel.com>
* Get default config based on the auto-detect CPU type (#1904)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Add export support for TEQ (#1910)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* update Gaudi CI baseline artifacts name (#1912)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Remove deprecated modules (#1872)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* fix CI docker container clean up issue (#1917)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* remove 1x docs (#1900)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Add `save`/`load` support for HQQ (#1913)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Co-authored-by: chen, suyue <suyue.chen@intel.com>
* Support PT2E save and load (#1918)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* implement TorchBaseConfig (#1911)
Signed-off-by: xin3he <xin3.he@intel.com>
* update documentation for 3x API (#1923)
Signed-off-by: chensuyue <suyue.chen@intel.com>
Signed-off-by: xin3he <xin3.he@intel.com>
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* fix typo in architecture diagram (#1924)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* Support woq Autotune (#1921)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Support absorb dict for awq (#1920)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Support LayerWise for RTN/GPTQ (#1883)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: chensuyue <suyue.chen@intel.com>
* update itrex ut test (#1929)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* add docstring for torch.quantization and torch.utils (#1928)
Signed-off-by: xin3he <xin3.he@intel.com>
* Integrate AutoRound v0.3 (#1925)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Integrate AutoRound v0.3 to 2x (#1926)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Enhance load_empty_model import (#1930)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Add doc for client usage (#1914)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* remove peft version limit (#1933)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Support xpu for ipex static quant (#1916)
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* Support calib_func on TF 3x API (#1934)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* 3.X API installation update (#1935)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Fix unused pkgs import (#1931)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Add docstring for PT2E and HQQ (#1937)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* add docstring for static quant and smooth quant (#1936)
* add docstring for static quant and smooth quant
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* format fix
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* update scan path
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* Update utility.py
---------
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
Co-authored-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* Update Example for Pytorch 3x Mixed Precision (#1882)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* add read permission token (#1942)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* Add docstring for WOQ&LayerWise (#1938)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: xinhe <xin3.he@intel.com>
* add docstring for mx quant (#1932)
Signed-off-by: Mengni Wang <mengni.wang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: xinhe <xin3.he@intel.com>
* Update for API 3.0 online doc (#1940)
Co-authored-by: ZhangJianyu <zhang.jianyu@outlook.com>
* Refine Pytorch 3x Mixed Precision Example (#1946)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* Update AutoRound commit version (#1941)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Update publish.yml (#1949)
* Update publish.yml
* Update publish.yml
* Update publish.yml (#1950)
* Update doc for client-usage and LWQ (#1947)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Add Docstring for TF 3x API and Torch 3x Mixed Precision (#1944)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* Update Examples for TF 3x API (#1901)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* Complement UT of calibration function for TF 3x API (#1945)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* Enable yolov5 Example for TF 3x API (#1943)
Signed-off-by: zehao-intel <zehao.huang@intel.com>
* add ipex xpu example to 3x API (#1948)
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* update 3x torch installation (#1957)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Add save/load for pt2e example (#1927)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Fix itrex qbits nf4/int8 training core dumped issue (#1954)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Signed-off-by: chensuyue <suyue.chen@intel.com>
* new previous results could not find all raise issues in CI model test (#1958)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Set low_gpu_mem_usage=False for AutoRound
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Bump tensorflow version (#1961)
Signed-off-by: dependabot[bot] <support@github.com>
* fix docs link (#1959)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* fix welcome.html link issue (#1962)
Co-authored-by: ZhangJianyu <zhang.jianyu@outlook.com>
* replenish docstring (#1955)
* replenish docstring
Signed-off-by: xin3he <xin3.he@intel.com>
* update Quantizer API docstring
Signed-off-by: xin3he <xin3.he@intel.com>
* Add docstring for auto accelerator (#1956)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* temporary remove torch/quantization and add it back after fp8 code is updated.
* Update config.py
---------
Signed-off-by: xin3he <xin3.he@intel.com>
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Co-authored-by: Yi Liu <106061964+yiliu30@users.noreply.github.com>
* add SDXL model example to INC 3.x (#1887)
* add SDXL model example to INC 3.x
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
* add evaluation script
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* add test script
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* minor fix
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* Update run_quant.sh
* add iter limit
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* modify test script
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* update json
Signed-off-by: chensuyue <suyue.chen@intel.com>
* add requirements
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* Update run_benchmark.sh
* Update sdxl_smooth_quant.py
* minor fix
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
---------
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
Signed-off-by: chensuyue <suyue.chen@intel.com>
Co-authored-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
Co-authored-by: chensuyue <suyue.chen@intel.com>
* example update for 3.x ipex sq (#1902)
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
* Fix `opt_125m_woq_gptq_int4_dq_ggml` issue (#1965)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* remove unnecessary CI (#1966)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Add version mapping between INC and Gaudi SW Stack (#1967)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* Add 3.x readme (#1971)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Fix broken link in docs (#1969)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* Cherry pick v1.17.0 (#1964)
* [SW-184941] INC CI, CD and Promotion
Change-Id: I60c420f9776e1bdab7bb9e02e5bcbdb6891bfe52
* [SW-183320]updated setup.py
Change-Id: I592af89486cb1d9e0b5197521c428920197a9103
* [SW-177474] add HQT FP8 porting code
Change-Id: I4676f13a5ed43c444f2ec68675cc41335e7234dd
Signed-off-by: Zhou Yuwen <zyuwen@habana.ai>
* [SW-189361] Fix white list extend
Change-Id: Ic2021c248798fce37710d28014a6d59259c868a3
* [SW-191317] Raise exception according to hqt config object
Change-Id: I06ba8fa912c811c88912987c11e5c12ef328348a
* [SW-184714] Port HQT code into INC
HQT lib content was copied as is under fp8_quant
Tests were copied to 3.x torch location
Change-Id: Iec6e1fa7ac4bf1df1c95b429524c40e32bc13ac9
* [SW-184714] Add internal folder to fp8 quant
This is a folder used for experiments,
not to be used by users
Change-Id: I9e221ae582794e304e95392c0f37638f7bce69bc
* [SW-177468] Removed unused code + cleanup
Change-Id: I4d27c067e87c1a30eb1da9df16a16c46d092c638
* Fix errors in regression_detection
Change-Id: Iee5318bd5593ba349812516eb5641958ece3c438
* [SW-187731] Save orig module as member of patched module
This allows direct usage of the original module methods,
which solves torch compile issue
Change-Id: I464d8bd1bacdfc3cd1f128a67114e1e43f092632
* [SW-190899] Install packages according to configuration
Change-Id: I570b490658f5d2c5399ba1db93f8f52f56449525
* [SW-184689] use finalize_calibration intrenaly for one step flow
Change-Id: Ie0b8b426c951cf57ed7e6e678c86813fb2d05c89
* [SW-191945] align requirement_pt.txt in gerrit INC with Github INC
Change-Id: If5c0dbf21bf989af37a8e29246e4f8760cd215ef
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* [SW-192358] Remove HQT reference in INC
Change-Id: Ic25f9323486596fa2dc6d909cd568a37ab84dd5e
* [SW-191415] update fp8 maxAbs observer using torch.copy_
Change-Id: I3923c832f9a8a2b14e392f3f4719d233a457702f
* [SW-184943] Enhance INC WOQ model loading
- Support loading huggingface WOQ model
- Abstract WeightOnlyLinear base class. Add INCWeightOnlyLinear and HPUWeighOnlyLinear subclasses
- Load woq linear weight module by module
- Save hpu format tensor to reuse it once load it again
Change-Id: I679a42759b49e1f45f52bbb0bdae8580a23d0bcf
* [SW-190303] Implement HPUWeightOnlyLinear class in INC
Change-Id: Ie05c8787e708e2c3559dce24ef0758d6c498ac41
* [SW-192809] fix json_file bug when instantiating FP8Config class
Change-Id: I4a715d0a706efe20ccdb49033755cabbc729ccdc
Signed-off-by: Zhou Yuwen <zyuwen@habana.ai>
* [SW-192931] align setup.py with github INC and remove fp8_convert
Change-Id: Ibbc157646cfcfad64b323ecfd96b9bbda5ba9e2f
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* [SW-192917] Update all HQT logic files with pre-commit check
Change-Id: I119dc8578cb10932fd1a8a674a8bdbf61f978e42
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* update docstring
Signed-off-by: yuwenzho <yuwen.zhou@intel.com>
* add fp8 example and document (#1639)
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* Update settings to be compatible with gerrit
* enhance ut
Signed-off-by: yuwenzho <yuwen.zhou@intel.com>
* move fp8 sample to helloworld folder
Signed-off-by: yuwenzho <yuwen.zhou@intel.com>
* update torch version of habana docker
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* update readme demo
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* update WeightOnlyLinear to INCWeightOnlyLinear
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* add docstring for FP8Config
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* fix pylint
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* update fp8 test scripts
Signed-off-by: chensuyue <suyue.chen@intel.com>
* delete deps
Signed-off-by: chensuyue <suyue.chen@intel.com>
* update container into v1.17.0
Signed-off-by: chensuyue <suyue.chen@intel.com>
* update docker version
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* update pt ut
Signed-off-by: chensuyue <suyue.chen@intel.com>
* add lib path
Signed-off-by: chensuyue <suyue.chen@intel.com>
* fix dir issue
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* update fp8 test scope
Signed-off-by: chensuyue <suyue.chen@intel.com>
* fix typo
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* update fp8 test scope
Signed-off-by: chensuyue <suyue.chen@intel.com>
* update pre-commit-ci
Signed-off-by: chensuyue <suyue.chen@intel.com>
* work around for hpu
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* fix UT
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* fix parameter
Signed-off-by: chensuyue <suyue.chen@intel.com>
* omit some test
Signed-off-by: chensuyue <suyue.chen@intel.com>
* update main page example to llm loading
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix autotune
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
---------
Signed-off-by: Zhou Yuwen <zyuwen@habana.ai>
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
Signed-off-by: yuwenzho <yuwen.zhou@intel.com>
Signed-off-by: chensuyue <suyue.chen@intel.com>
Co-authored-by: yan tomsinsky <ytomsinsky@habana.ai>
Co-authored-by: Ron Ben Moshe <rbenmoshe@habana.ai>
Co-authored-by: Uri Livne <ulivne@habana.ai>
Co-authored-by: Danny Semiat <dsemiat@habana.ai>
Co-authored-by: smarkovichgolan <smarkovich@habana.ai>
Co-authored-by: Dudi Lester <dlester@habana.ai>
* update main page (#1973)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* fix online doc search issue (#1975)
Co-authored-by: ZhangJianyu <zhang.jianyu@outlook.com>
* bump main version into v3.1 (#1974)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* update readme for fp8 (#1979)
Signed-off-by: xinhe3 <xinhe3@habana.ai>
* Skip some tests for torch 2.4 (#1981)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Fix UT env and upgrade torch to 2.4.0 (#1978)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* support gptq `true_sequential` and `quant_lm_head` (#1977)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* update installation and ci test for 3x api (#1991)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* add hasattr check for torch fp8 dtype (#1985)
Signed-off-by: xin3he <xin3.he@intel.com>
* add quantize, save, load function for transformers-like api (#1986)
Signed-off-by: changwangss <chang1.wang@intel.com>
* Update installation_guide.md (#1989)
Correct typo in installation doc
* update 3x pt binary build (#1992)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* add per_channel_minmax (#1990)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Remove the save of gptq config (#1993)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Add recent publications (#1995)
* add recent publications
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* update total count
Signed-off-by: Huang, Tai <tai.huang@intel.com>
---------
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* update docker image prune rules (#2003)
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Support transformers-like api for woq quantization (#1987)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Wang, Chang <chang1.wang@intel.com>
* add INC_FORCE_DEVICE introduction (#1988)
* add INC_FORCE_DEVICE introduction
Signed-off-by: xin3he <xin3.he@intel.com>
* Update PyTorch.md
* Update PyTorch.md
* Update docs/source/3x/PyTorch.md
Co-authored-by: Yi Liu <yi4.liu@intel.com>
* rename to INC_TARGET_DEVICE
Signed-off-by: xin3he <xin3.he@intel.com>
---------
Signed-off-by: xin3he <xin3.he@intel.com>
Co-authored-by: Yi Liu <yi4.liu@intel.com>
* Replace FORCE_DEVICE with INC_TARGET_DEVICE [transformers] (#2005)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* enable auto_round format export (#2002)
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* remove accelerate version in unit test (#2007)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* add repack_awq_to_optimum_format function (#1998)
Signed-off-by: changwangss <chang1.wang@intel.com>
* Update auto_round requirements for transformers example (#2013)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* add pad_to_buckets in evaluation for hpu performance (#2011)
* add pad_to_buckets in evaluation for hpu performance
---------
Signed-off-by: xin3he <xin3.he@intel.com>
* Update model accuracy (#2006)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* fix xpu device set weight and bias (#2010)
Signed-off-by: changwangss <chang1.wang@intel.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
* Add transformers-like api doc (#2018)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Adapt transformers 4.45.1 (#2019)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: changwangss <chang1.wang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* add autoround EMNLP24 to pub list (#2014)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* Fix transformers rtn layer-wise quant (#2008)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Remove itrex dependency for 3x example (#2016)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
* add transformers-like api link in readme (#2022)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* Add woq examples (#1982)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
* remove ITREX unit test CI (#2021)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Support quant procedure on XPU (#2026)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Support generation search for transformers examples (#2029)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Remove itrex dependency for 2x example (#2024)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Update the PT2E CV example (#2032)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Cherry pick Habana software 1.18.0 update (#2025)
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Signed-off-by: Yi Liu <yiliu4@habana.ai>
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
Signed-off-by: chensuyue <suyue.chen@intel.com>
Co-authored-by: yan tomsinsky <ytomsinsky@habana.ai>
Co-authored-by: Uri Livne <ulivne@habana.ai>
Co-authored-by: Dudi Lester <dlester@habana.ai>
Co-authored-by: Danny <dsemiat@habana.ai>
Co-authored-by: Tomer Gafni <tgafni@habana.ai>
Co-authored-by: Eran Geva <egeva@habana.ai>
Co-authored-by: Daniel Ohayon <danielohayon444@gmail.com>
Co-authored-by: Roi Tiefenbrunn <rtiefenbrunn@habana.ai>
Co-authored-by: Kamil Felskowski <kfelskowskix@habana.ai>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* update gaudi version mapping table for v3.1 (#2030)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
Co-authored-by: chen, suyue <suyue.chen@intel.com>
* fix broken link to FP8 example (#2034)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* add back missing image (#2035)
Signed-off-by: xin3he <xin3.he@intel.com>
* Add vlm examples, bugfix (#2012)
* add VLM examples
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* bugfix, add utils
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix docstring issues
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* bugfix
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* refine examples
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* fix scan issue
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* refine shell
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* refine scripts & requirements
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* typofix
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* refine docs
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* set attn_implementation for Phi3-vision
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* refine phi3 example
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix code coverage
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* update config
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* refine shells, docs and example. enable qwen2-vl quantization
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fix ci
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* fix EOF error
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* update qwen dir
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* refine shell, add llama3.2 inference to doc
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* bugfix
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* bugfix
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* bugfix
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* refine eval shell
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* fix eval device issue
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* refine eval dtype
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
---------
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
* remove autoround limit (#2036)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Adapt autoround format (#2038)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* remove transformers import from utility (#2045)
* remove transformers import from utility
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* bugfix
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* fixtypos
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
---------
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* add buckets setting for lm_eval (#2044)
* add buckets setting for lm_eval
Signed-off-by: xinhe3 <xinhe3@habana.ai>
* clear graph cache to avoid OOM
Signed-off-by: xinhe3 <xinhe3@habana.ai>
---------
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Co-authored-by: xinhe3 <xinhe3@habana.ai>
* Enhance example for HPU performance (#2043)
* Enhance example for HPU performance
Signed-off-by: xinhe3 <xinhe3@habana.ai>
* Update run_clm_no_trainer.py
* remove wikitext to avoid oom for llama2-7b bs=8
* remove wikitext
Signed-off-by: xinhe3 <xinhe3@habana.ai>
---------
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Co-authored-by: xinhe3 <xinhe3@habana.ai>
* remove useless code in setup.py (#2046)
* Update the default PT2E config (#2041)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Support non-contiguous weight saving (#2049)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* fix GPTQ oom issue on HPU (#2042)
* fix GPTQ oom issue on HPU
Signed-off-by: xinhe3 <xinhe3@habana.ai>
---------
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Co-authored-by: xinhe3 <xinhe3@habana.ai>
* fix bug and update readme (#2051)
* fix bug and update readme
---------
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Co-authored-by: xinhe3 <xinhe3@habana.ai>
* Support safetensors loading for layerwise (#2047)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Enhance WOQ example Readme and help (#2053)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: xinhe <xin3.he@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* improve optimum-habana available check (#2054)
Signed-off-by: changwang <changwang@habana.ai>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Fixed CI IPEX version (#2061)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Update torch config kwargs (#2055)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Support client `use_layer_wise` setting (#2048)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Check autoround before import it (#2062)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Delete fp8_quant/scripts/regression_detection directory (#2059)
A missed change when cherry-picking Habana software 1.18.0
* Make PatchedVLLMKVCache resiliant to forward API changes (#2067)
Change-Id: I33fad5c3e80e017099f300782809f24669765d42
Co-authored-by: Konrad Zawora <kzawora@habana.ai>
* Fix glm-4-9b oom issue on BMG
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Update recipes & Bump version to 3.2 (#2037)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Docs: Add customer defined calibration and update docker run (#2057)
Signed-off-by: fengding <feng1.ding@intel.com>
* Adapt torch and ipex 2.5 (#2066)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
* Enhance `TBB` check (#2068)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Fix the PT2E UT (#2071)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* Support gptq layerwise on client (#2069)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Adapt autoround v0.4 (#2073)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Ensure that mul operators with shared initializer will not be absorbed in SmoothQuant (#2063)
Signed-off-by: duansheng.liu <44742794+duanshengliu@users.noreply.github.com>
* Integrate AutoRound v0.4 [3x] (#2072)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Update CI framework versions and README badge for release 3.1.1 (#2058)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Remove the examples force required torch 1.13.1 (#2074)
* remove alexnet_fashion_mnist notebook
Signed-off-by: chensuyue <suyue.chen@intel.com>
* remove rnnt in pytorch examples
Signed-off-by: chensuyue <suyue.chen@intel.com>
---------
Signed-off-by: chensuyue <suyue.chen@intel.com>
* Fix truthfulqa task evaluation issue
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Add required library for ONNX example (#2078)
* Add required library for ONNX example
* Update requirements.txt
* support autoround new API for VLM (#2075)
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* add import check (#2076)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* Update utility.py (#2079)
* Add gptq known issue (#2080)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Fix sdxl `q_unet` config (#2081)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* Fixed the PT2E LLM example (#2082)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* fix dlrm when using incbench (#2084)
Signed-off-by: Xin He <xinhe3@habana.ai>
* add mapping for v3.2 (#2085)
Signed-off-by: Huang, Tai <tai.huang@intel.com>
* [SW-192753] unify StaticQuantConfig and FP8Config
Change-Id: I2fe09ba4c575810a5b130268d63b9eee926bdf08
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-200124] Set Scalar as default scale format + Compatibility check
Set ScaleFormat.SCALAR as default value of 'scale_format'
Add reduction of 'scale_format' to 'CONST' value if using a PCQ scale_format or fake_quant
Add test to show Scalar models aren't giving wrong outputs
Fix fakequant test as it is problematic use of 'hpu_initialize' and should be fixed in SW-202697
Change-Id: I43ff4900e9e02ce7f50edcdbb19a28f4f615ef9c
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-201679] support unit_scales for FuseMoE
Change-Id: I02a63332bc09f1f6cdc3f133dd5f58829fcbad5a
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-203698] Add log for converting prepared model
Change-Id: I1464f11bbab27d9041c9ba6f448e5ae6fa43bc2d
Signed-off-by: Mengni Wang <mewang@habana.ai>
* [SW-199737] Measurement dump improvements
Add _validate_dump_path to make sure dump dir is writable and backup measurements
Change-Id: Ib64abe772b4c309bbf04de89477cde92ea47ade4
* [SW-203452] Fixing and temp skipping G3 unittests
Change-Id: Iafa4a6a8577724bd8a86581bfe38d3269dab2ea2
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-195965] [GPTQ] INC load model loads model in fp32 only
Change-Id: I597d19273786c0c169ad952ebe5a357274e358dc
Signed-off-by: xinhe3 <xinhe3@habana.ai>
* [SW-204016] Enable scale calculation with disk offload in INC
-move calculating scales and quantization config info during the module
patching loop as the weights there guaranteed to be on cpu.
Change-Id: Ifb2de4e67c1b36c611dcc50b4cd14731b0336c50
* [SW-202614] Llama70b int4 gptq with INC load flow - getting host OOM
Change-Id: Id1797371bb136502d89c4e8d17abcac1eaac4534
Signed-off-by: xinhe3 <xinhe3@habana.ai>
* [SW-199823] [HQT] fix INC one-step quantization API workflow
1. fix test_fp8_static_quant.py::TestFP8StaticQuant::test_one_step_quant_cv failure by deepcoping forward function in common.py
2. fix config.py: Object of type dict_keys is not JSON serializable by converting it to list
3. fix download issue of UT by using local tiny_gptj.json
Change-Id: I2ad3eac411e8fca9d88a021f6a5b9594e6c75ae9
Signed-off-by: xinhe3 <xinhe3@habana.ai>
* [SW-202617] vllm mixtral MoE quant and measure using forward call
Change-Id: I919f1e3597b6c95c3fc60db78ac9c0c06242b416
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-200092] Allow fsdpa and softmax to use scalar scales in INC
Change-Id: Ieba4c74c18624fb0c5fce6321671d6f4eb2b8c93
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-205363] Update _load_state_dict_into_meta_model
Update _load_state_dict_into_meta_model to compatible with
Transformer 4.45 release
Change-Id: Ib5d8ca777d38c7ae225b7174a886b333b6246ab1
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-184948] INC Q/DQ optimization, included conv2d, kv_cache, fsdpa,
softmax and other operators.
Change-Id: I920f8ad85b3493f1bd4bbe770533343e214fc2d1
Signed-off-by: changwang <changwang@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-198585] Fix typo causing PatchedVLLMKVCache error
Change-Id: Iafdcc935f702bc4756e2ba89935becb3bc47a728
* [SW-199208] QDQ Refactor for Registering Patched Modules, Scaling Methods, and Observers
1. Extension APIs
- `PatchedModuleBase` , `register_patched_module`
- `ScalingMethodBase`, `register_scaling_methods`
- `ObserverBase` ``register_observer`, `register_module_config_for_observer`
Related files:
- fp8_quant/patched_module_base.py
- fp8_quant/observer_base.py
- fp8_quant/_core/measure.py
- test_register_apis.py
2. Device-agnostic Patching
- Replaced `hpu` with `cur_accelerator.name()`
- Replaced `htcore.mark_step()` with `cur_accelerator.synchronize()`
- Removed `torch.device("hpu")` under observers and scaling method
- Updated `hpu_accelerator.synchronize()` to `htcore.mark_step()` + `torch.hpu.synchronize()`
Change-Id: I83c6de928a991ed2c1b3b434d372f49e095c38d3
Signed-off-by: Yi Liu <yiliu4@habana.ai>
Co-authored-by: Mengni Wang <mewang@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-203389] scalars scales doesn't provide dtype attribution
Change-Id: I4e40dc9b2d9cb65bc9e49571cd57a9ab030f5d7b
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-199208] fix ModuleInfo conversion issue
Change-Id: Ib6c35e1623dda3e470e569defccd607a18b43312
* [SW-200168] Enable working with G2 HW scales on G3
Change-Id: I17f71540eb78e828f01f1a11c8b233d60951293e
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-203389] fix get_scale_dtype to support PCQ scales
Change-Id: I923ace405a0f751a2e5a0a3aadb7abbb401a6c44
* [SW-199719] reduce PCQ scales memory usage
removed persistent full weight scales during PCQ quantization
instead we are keeping only the input and output channels scales
creating temporary full scale tensor on input quant Op call
since the full scale tensor is the same size as the orig bf16 weight
keeping all full scales persistently and the quntized weights will
result a quantized model that uses more memory than the unquantized.
Change-Id: Idc91c5ac8b9cfea2e2a3ad053cb4dc5464cff776
* [SW-206112] INC Q/DQ improvement - use Q/DQ ops
Change-Id: Ib03ea8744aa2cca8b606754c45944840da1c3898
Signed-off-by: changwang <changwang@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-206693] Convert conv2d_fp8 params to list if necessary
It's needed for the new approach to dynamic shapes in PT2.5.
Change-Id: I8d5e620153970b210675459e3d6aecad8ca7cbde
* [SW-207411] Add catch for OSError in _validate_dump_path
Change-Id: I82bae184257f3da982877b3797f2ee8b40a573c8
* [SW-207328] remove accuracy check due to random issue
Change-Id: Ifbd985c31c3755b6ab353ef8fa45e911dd75d688
Signed-off-by: xinhe3 <xinhe3@habana.ai>
* [SW-207559] Folder layout refactoring and cleanup (phase 1)
Change-Id: Ic9bffd2b7477d4530b4e2a5e411760a731efb84b
Signed-off-by: Yi Liu <yiliu4@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-193262] INC multi device save/load CP design in fp8 (#5)
Signed-off-by: Xin <xin3.he@intel.com>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-208521] one-step quantization got double memory usage (#3)
* [SW-208521] one-step quantization got double memory usage
Signed-off-by: Xin <xin3.he@intel.com>
* [SW-208789] Support quantizing FP16 model to FP8 (#15)
Since layer-wise is using memory mapping from disk, the model could be fp16 as it saved on disk, for example, llama2-7b.
We need to add logic to support this case to make sure layer-wise works well.
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-205959] Update _load_state_dict_into_meta_model for model with bias (#7)
Signed-off-by: Xin <xin3.he@intel.com>
* [SW-208700] release bf16 model memory on HPU in one-step quantization (#14)
Signed-off-by: Xin <xin3.he@intel.com>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-197077] refactoring maxabs scales and adding arbitrary scales. (#12)
* [SW-197077] refactoring maxabs scales and adding arbitrary scales.
Change-Id: I2c35cf925b6b21983f1770db7d35e14f3d7d3e47
* [SW-197077] refactoring scale:
fix atol
Change-Id: I1c99ddd9ade679286988e7d8a96338b32c0ddc07
* [SW-197077] adding arbitrary scales
* Skip autoround test for HPU (#19)
Change-Id: I6dc9724389c16a05252370b9e09a1db80bc8d696
Signed-off-by: Yi Liu <yiliu4@habana.ai>
Co-authored-by: Yi Liu <yiliu4@habana.ai>
* [SW-199728] [DeepSpeed] Buffers initialized by model are not correct … (#16)
* [SW-199728] [DeepSpeed] Buffers initialized by model are not correct after tensor parallel
---------
Signed-off-by: Xin <xin3.he@intel.com>
Co-authored-by: Danny Semiat <dsemiat@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-208151] CD 1.19.0 - PT Docker - test_quantization No module named… (#33)
* [SW-209256] fix GPTQ oom issue on HPU (#2042) (#20)
* fix GPTQ oom issue on HPU (#2042)
---------
Signed-off-by: Xin <xin3.he@intel.com>
Co-authored-by: xinhe3 <xinhe3@habana.ai>
* [SW-208151] CD 1.19.0 - PT Docker - test_quantization No module named 'safetensors'
Signed-off-by: Xin <xin3.he@intel.com>
---------
Signed-off-by: Xin <xin3.he@intel.com>
Co-authored-by: xinhe3 <xinhe3@habana.ai>
Co-authored-by: Danny Semiat <dsemiat@habana.ai>
* [SW-207748] Support Auto-round on HPU (#25)
Signed-off-by: Yi Liu <yiliu4@habana.ai>
Co-authored-by: Yi Liu <yiliu4@habana.ai>
* [SW-209878] Increase threshold to avoid random error in test_layer_wise.py (#36)
Signed-off-by: Xin He <xinhe3@habana.ai>
Co-authored-by: Xin He <xinhe3@habana.ai>
* [SW-207579] support load vLLM compatible FP8 model (#18)
Support load vLLM compatible FP8 model, both G2 and G3, both single card and multi-cards.
---------
Signed-off-by: changwang <changwang@habana.ai>
* [SW-207451] Implement block-wise calibration for LLM (#41)
* [SW-207451] Implement block-wise calibration for LLM
---------
Signed-off-by: Xin <xin3.he@intel.com>
Co-authored-by: Xin He <xinhe3@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-208986] fix save&load bug (#40)
* [SW-208986] fix save&load bug
---------
Signed-off-by: Xin He <xinhe3@habana.ai>
Co-authored-by: Xin He <xinhe3@habana.ai>
* [SW-207748] Add Auto-round Example (#42)
* add autoround hpu example
Change-Id: Ibd537f4667c7c077160427722a5eca2c721aa5cd
Signed-off-by: Yi Liu <yiliu4@habana.ai>
* add requirements
Change-Id: I77a95ec05e41247db9903e8622c31f05259ca365
Signed-off-by: Yi Liu <yiliu4@habana.ai>
---------
Signed-off-by: Yi Liu <yiliu4@habana.ai>
Co-authored-by: Yi Liu <yiliu4@habana.ai>
Co-authored-by: Uri Livne <ulivne@habana.ai>
Signed-off-by: Xin He <xinhe3@habana.ai>
* [SW-197077] fix bug (#47)
* [SW-210541] loading for fused_sdpa requires additional amax scale (#51)
Signed-off-by: Xin He <xinhe3@habana.ai>
Co-authored-by: Xin He <xinhe3@habana.ai>
* fix PatchedLoRACompatibleLinear init (#65)
Signed-off-by: changwangss <changwang@habana.ai>
* align files with v1.19.0 in fp8_quant folder
Signed-off-by: Xin He <xinhe3@habana.ai>
* fix missing SaveLoadFormat
Signed-off-by: Xin He <xinhe3@habana.ai>
* align and fix config after cherry-pick
Signed-off-by: Xin He <xinhe3@habana.ai>
* Implicit relative imports is abandoned
Signed-off-by: Xin He <xinhe3@habana.ai>
* fix config issue blocking CI
Signed-off-by: Xin He <xinhe3@habana.ai>
* remove synchronize for `pack_unpack_tensor_with_numpy` (#2070)
* remove pack&unpack synchronize
---------
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* stop auto-fix of pre-commit
Signed-off-by: Xin He <xinhe3@habana.ai>
* update autoround example for release test
Signed-off-by: xin3he <xin3.he@intel.com>
* fix AWQ&TEQ loading due to input scale
Signed-off-by: xin3he <xin3.he@intel.com>
* fix HQQ state_dict loading caused by [SW-195965]
Signed-off-by: xin3he <xin3.he@intel.com>
* use per_channel as default config (#2091)
Signed-off-by: yiliu30 <yi4.liu@intel.com>
* workaround transformers issue in version 4.47.0 (#2092)
* workaround transformers issue in version 4.47.0
Signed-off-by: xin3he <xin3.he@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* Refactor FP8 pytest script (#2089)
* Refactor FP8 pytest script
---------
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* update ci scan scope
Signed-off-by: chensuyue <suyue.chen@intel.com>
* [SW-210500] [Optimum-Habana] [Regression] [fp8] [INC] No generated text for llava models [llava-1.5-7b-hf] [llava-1.5-13b-hf ] (#54)
Signed-off-by: Xin He <xinhe3@habana.ai>
Co-authored-by: Xin He <xinhe3@habana.ai>
* [SW-213236] resolve CPU mem issue in CI (#76)
Signed-off-by: Xin He <xinhe3@habana.ai>
Co-authored-by: Xin He <xinhe3@habana.ai>
* recover pre-commit
Signed-off-by: Xin He <xinhe3@habana.ai>
* [pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
* Fix `is_sharded` setting for loading quant model (#2094)
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
* fix error message for different python version (#2099)
Signed-off-by: changwangss <changwang@habana.ai>
* fix UT of RTN on HPU (#2098)
Signed-off-by: xin3he <xin3.he@intel.com>
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* fix device issue during calibration (#2100)
Signed-off-by: Xin He <xinhe3@habana.ai>
* fix woq example and update document for v1.19.0 (#2097)
Signed-off-by: xin3he <xin3.he@intel.com>
* Refactor version import paths to common module (#2095)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* update CI gaudi-docker to 1.19.0 (#2096)
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
* fix device mapping issue of llama gptq (#2101)
Signed-off-by: Xin He <xinhe3@habana.ai>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
* remove fix_measurements.py. exists with a different name - postprocessing_vllm_measurements.py
* fix merge
* remove unused imported functions with wrong path
* change envar requested value from 1 to true
---------
Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>
Signed-off-by: zehao-intel <zehao.huang@intel.com>
Signed-off-by: xin3he <xin3.he@intel.com>
Signed-off-by: Kaihui-intel <kaihui.tang@intel.com>
Signed-off-by: Sun, Xuehao <xuehao.sun@intel.com>
Signed-off-by: chensuyue <suyue.chen@intel.com>
Signed-off-by: yiliu30 <yi4.liu@intel.com>
Signed-off-by: He, Xin3 <xin3.he@intel.com>
Signed-off-by: changwangss <chang1.wang@intel.com>
Signed-off-by: Huang, Tai <tai.huang@intel.com>
Signed-off-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
Signed-off-by: Mengni Wang <mengni.wang@intel.com>
Signed-off-by: dependabot[bot] <support@github.com>
Signed-off-by: Zhou Yuwen <zyuwen@habana.ai>
Signed-off-by: xinhe3 <xinhe3@hababa.ai>
Signed-off-by: yuwenzho <yuwen.zhou@intel.com>
Signed-off-by: xinhe3 <xinhe3@habana.ai>
Signed-off-by: Zhang, Weiwei1 <weiwei1.zhang@intel.com>
Signed-off-by: Yi Liu <yiliu4@habana.ai>
Signed-off-by: changwang <changwang@habana.ai>
Signed-off-by: fengding <feng1.ding@intel.com>
Signed-off-by: duansheng.liu <44742794+duanshengliu@users.noreply.github.com>
Signed-off-by: Xin He <xinhe3@habana.ai>
Signed-off-by: Mengni Wang <mewang@habana.ai>
Signed-off-by: Xin <xin3.he@intel.com>
Signed-off-by: changwangss <changwang@habana.ai>
Co-authored-by: Zixuan Cheng <110808245+violetch24@users.noreply.github.com>
Co-authored-by: xinhe <xin3.he@intel.com>
Co-authored-by: zehao-intel <zehao.huang@intel.com>
Co-authored-by: Kaihui-intel <kaihui.tang@intel.com>
Co-authored-by: Sun, Xuehao <xuehao.sun@intel.com>
Co-authored-by: chen, suyue <suyue.chen@intel.com>
Co-authored-by: Yi Liu <106061964+yiliu30@users.noreply.github.com>
Co-authored-by: Dina Suehiro Jones <dina.s.jones@intel.com>
Co-authored-by: Wang, Chang <chang1.wang@intel.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Huang, Tai <tai.huang@intel.com>
Co-authored-by: violetch24 <zixuan@aia-sdp-spr-117706.jf.intel.com>
Co-authored-by: Wang, Mengni <mengni.wang@intel.com>
Co-authored-by: Neo Zhang Jianyu <jianyu.zhang@intel.com>
Co-authored-by: ZhangJianyu <zhang.jianyu@outlook.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: yan tomsinsky <ytomsinsky@habana.ai>
Co-authored-by: Ron Ben Moshe <rbenmoshe@habana.ai>
Co-authored-by: Uri Livne <ulivne@habana.ai>
Co-authored-by: Danny Semiat <dsemiat@habana.ai>
Co-authored-by: smarkovichgolan <smarkovich@habana.ai>
Co-authored-by: Dudi Lester <dlester@habana.ai>
Co-authored-by: Yi Liu <yi4.liu@intel.com>
Co-authored-by: WeiweiZhang1 <weiwei1.zhang@intel.com>
Co-authored-by: Tomer Gafni <tgafni@habana.ai>
Co-authored-by: Eran Geva <egeva@habana.ai>
Co-authored-by: Daniel Ohayon <danielohayon444@gmail.com>
Co-authored-by: Roi Tiefenbrunn <rtiefenbrunn@habana.ai>
Co-authored-by: Kamil Felskowski <kfelskowskix@habana.ai>
Co-authored-by: xinhe3 <xinhe3@habana.ai>
Co-authored-by: Konrad Zawora <kzawora@habana.ai>
Co-authored-by: feng-intel <110514170+feng-intel@users.noreply.github.com>
Co-authored-by: duanshengliu <44742794+duanshengliu@users.noreply.github.com>
Co-authored-by: Mengni Wang <mewang@habana.ai>
Co-authored-by: Jimin Ha <jha@habana.ai>
Co-authored-by: changwang <changwang@habana.ai>
Co-authored-by: Yi Liu <yiliu4@habana.ai>
Co-authored-by: Amadeusz Skrzypczak <askrzypczak@habana.ai>
Co-authored-by: Linoy Buchnik <linoybu@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Type of Change
request from optimum-intel huggingface/optimum-intel#1018 (comment)
to make
load_empty_modelhave the same input kwargs withAutoModelForCausalLM.from_pretrainedDescription
load_empty_function couldn't pass the **loading_kwargs from
AutoModelForCausalLM.from_pretrained, because we initialize the config first, and then initialize the model in load_empty_model function. some loading_kwargs like "subfolder", "revision" doesn't be accepted byAutoConfig.from_pretrainedkwargs.neural-compressor/neural_compressor/torch/utils/utility.py
Line 356 in e269660
when the **loading_kwargs passed, config initialization will raise error like the following.
Expected Behavior & Potential Risk
the expected behavior that triggered by this PR
How has this PR been tested?
how to reproduce the test (including hardware information)
Dependency Change?
any library dependency introduced or removed