mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-07-29 20:14:01 +00:00
459c750b03
### Description <!-- Describe your changes. --> 1. Make JBLAS codes an external module of ORT. 2. Move q4 gemm code to contrib_ops. 3. Update template kernel library to v0.1 release. ### Motivation and Context <!-- - Why is this change required? What problem does it solve? - If it fixes an open issue, please link to the issue here. --> We found that the current LLM model performance is far below our expectations. Here is some performance data collected on Mistral-7B model with Xeon-8480: 8 threads | prompt length=32 past_len=32 | prompt length=1 past_len=32 -- | -- | -- ORT-main | 1220ms | 263ms Neural-speed | 564ms | 87ms ORT-this PR|597ms|120ms Although `Neural-speed` and `ORT-this PR` use the same int4 kernel code, there is a 33ms(87ms vs. 120ms) latency gap between the two frameworks. Through some statistics analysis, the summary latency of `MatMulNBits` is 86.7ms The summary latency of all int4 GEMMs in `Neural-speed` is 84.8ms. So other OPs introduce an extra 30ms latency. The performance of MatMulNBits in this PR meets our expectations. ### Remain Issues 1. For hybrid CPUs, like core 12900K, the ONNXRuntime thread pool uses TaskGranularityFactor to scale its number of threads. This is not expected in our code design. It may slow down the hybrid CPU performance by 30~40%. 2. Prepack uses a single thread which is very slow to init a session. 3. MatMulNBits with zero points will fall through to COMP_FP32 even accuracy_level=4. Our COMP_INT8 IGemmCore with zero points process is not optimized for now. It will be updated in the future. So, for an int4 model with zero points, whether the accuracy_level is 0 or 4 will be no difference.
57 lines
No EOL
6.6 KiB
Text
57 lines
No EOL
6.6 KiB
Text
#Name;Url;SHA1
|
|
#This is a CSV file.
|
|
#The columns are separated by ";" because a list in cmake is just a ";" separated group of strings.
|
|
#Names should be in lower case. They will be used as variable names in cmake.
|
|
#URLs can be either https URLs or local file paths in cmake-style(directory separator is a forward slash character).
|
|
#SHA1 hashes can be generated by running sha1sum command on linux. PowerShell can also be used:
|
|
# (Get-FileHash -Algorithm SHA1 <filename>).Hash.ToLower()
|
|
#If you need to change abseil's version to a different one, you may also want to update external\abseil-cpp.natvis
|
|
#since the file contains a version string: "lts_20230802". However, the file is for debugging purposes only and would
|
|
#not affect built binaries.
|
|
#
|
|
# NOTE: You must run deps_update_and_upload.py and generate_cgmanifest.py when ready to test your changes in a CI.
|
|
# See https://microsoft.sharepoint.com/teams/ONNX2/_layouts/OneNote.aspx?id=%2Fteams%2FONNX2%2FShared%20Documents%2FNotebooks%2FONNX%20Ecosystem%20Team%20Notebook&wd=target%28Development.one%7C63D3AB47-51D1-4A62-9965-66882234BD44%2FAdd%20or%20update%20a%20dependency%20in%20deps.txt%7C0E9ED71D-89D5-40FA-B05F-C0123289C591%2F%29
|
|
#
|
|
abseil_cpp;https://github.com/abseil/abseil-cpp/archive/dcd5bd5fd593e31465af3d9ef291d26c646b0a4f.zip;6cc204586014e189f5c0fe3274f83162fa7c700c
|
|
cxxopts;https://github.com/jarro2783/cxxopts/archive/3c73d91c0b04e2b59462f0a741be8c07024c1bc0.zip;6c6ca7f8480b26c8d00476e0e24b7184717fe4f0
|
|
date;https://github.com/HowardHinnant/date/archive/refs/tags/v3.0.1.zip;2dac0c81dc54ebdd8f8d073a75c053b04b56e159
|
|
dlpack;https://github.com/dmlc/dlpack/archive/refs/tags/v0.6.zip;4d565dd2e5b31321e5549591d78aa7f377173445
|
|
# This Eigen commit id matches the eigen archive being consumed from https://gitlab.com/libeigen/eigen/-/archive/3.4/eigen-3.4.zip
|
|
# prior to the 3.4.1 RC changing the bits and invalidating the hash.
|
|
# it contains changes on top of 3.4.0 which are required to fix build issues.
|
|
# Until the 3.4.1 release this is the best option we have.
|
|
# Issue link: https://gitlab.com/libeigen/eigen/-/issues/2744
|
|
eigen;https://gitlab.com/libeigen/eigen/-/archive/e7248b26a1ed53fa030c5c459f7ea095dfd276ac/eigen-e7248b26a1ed53fa030c5c459f7ea095dfd276ac.zip;be8be39fdbc6e60e94fa7870b280707069b5b81a
|
|
flatbuffers;https://github.com/google/flatbuffers/archive/refs/tags/v1.12.0.zip;ba0a75fd12dbef8f6557a74e611b7a3d0c5fe7bf
|
|
fp16;https://github.com/Maratyszcza/FP16/archive/0a92994d729ff76a58f692d3028ca1b64b145d91.zip;b985f6985a05a1c03ff1bb71190f66d8f98a1494
|
|
fxdiv;https://github.com/Maratyszcza/FXdiv/archive/63058eff77e11aa15bf531df5dd34395ec3017c8.zip;a5658f4036402dbca7cebee32be57fb8149811e1
|
|
google_benchmark;https://github.com/google/benchmark/archive/refs/tags/v1.7.0.zip;e97c368b176e8614e3f1bf13dd9abcf6a7ad9908
|
|
google_nsync;https://github.com/google/nsync/archive/refs/tags/1.26.0.zip;5e7c00ef6bf5b787386fc040067903ec774e2752
|
|
googletest;https://github.com/google/googletest/archive/530d5c8c84abd2a46f38583ee817743c9b3a42b4.zip;5e3a61db2aa975cfd0f97ba92c818744e7fa7034
|
|
googlexnnpack;https://github.com/google/XNNPACK/archive/0da379fc4808f9601faef392352018c741c0f297.zip;663883491e380b628e0a5b162b5f2658032fae73
|
|
json;https://github.com/nlohmann/json/archive/refs/tags/v3.10.5.zip;f257f8dc27c5b8c085dc887b40cddd18ae1f725c
|
|
microsoft_gsl;https://github.com/microsoft/GSL/archive/refs/tags/v4.0.0.zip;cf368104cd22a87b4dd0c80228919bb2df3e2a14
|
|
microsoft_wil;https://github.com/microsoft/wil/archive/refs/tags/v1.0.230629.1.zip;e4a542a323c070376f7c2d1973d0f7ddbc1d2fa5
|
|
mimalloc;https://github.com/microsoft/mimalloc/archive/refs/tags/v2.1.1.zip;d5ee7d34223d0567892db5179849939c8769dc41
|
|
mp11;https://github.com/boostorg/mp11/archive/refs/tags/boost-1.82.0.zip;9bc9e01dffb64d9e0773b2e44d2f22c51aace063
|
|
onnx;https://github.com/onnx/onnx/archive/refs/tags/v1.15.0.zip;54c3f960a0541c5d8d3e60c2933e11f5d3688a11
|
|
#use the commit of supporting all the plugins and TRT 8.6-GA (https://github.com/onnx/onnx-tensorrt/commit/0462dc31ae78f48744b6141ae376df1f96d3f459)
|
|
onnx_tensorrt;https://github.com/onnx/onnx-tensorrt/archive/a43ce67187bab219520fd80f21af8bbd4354bc8c.zip;572535aefef477050f86744dfab1fef840198035
|
|
protobuf;https://github.com/protocolbuffers/protobuf/archive/refs/tags/v21.12.zip;7cf2733949036c7d52fda017badcab093fe73bfa
|
|
protoc_win64;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win64.zip;b4521f7ada5b260380f94c4bd7f1b7684c76969a
|
|
protoc_win32;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-win32.zip;3688010318192c46ce73213cdfb6b3e5656da874
|
|
protoc_linux_x64;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-linux-x86_64.zip;338462004aa5be9fba45b35b5b4be43f69b47a90
|
|
protoc_linux_x86;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-linux-x86_32.zip;61fdbe7d6360e065ec6fea23bca2cca673115fb8
|
|
protoc_linux_aarch64;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-linux-aarch_64.zip;df9d45470b0b8cf939dd2f0ec6b88e9cafc4d617
|
|
protoc_mac_universal;https://github.com/protocolbuffers/protobuf/releases/download/v21.12/protoc-21.12-osx-universal_binary.zip;23710c3d1c2036d8d65a6a22234372fa2d7af9ef
|
|
psimd;https://github.com/Maratyszcza/psimd/archive/072586a71b55b7f8c584153d223e95687148a900.zip;1f5454b01f06f9656b77e4a5e2e31d7422487013
|
|
pthreadpool;https://github.com/Maratyszcza/pthreadpool/archive/4fe0e1e183925bf8cfa6aae24237e724a96479b8.zip;07a0aa91dd9bf86f31b95497e00f31d8a261a4bd
|
|
pybind11;https://github.com/pybind/pybind11/archive/refs/tags/v2.10.1.zip;769b6aa67a77f17a770960f604b727645b6f6a13
|
|
pytorch_cpuinfo;https://github.com/pytorch/cpuinfo/archive/959002f82d7962a473d8bf301845f2af720e0aa4.zip;85da3caa60eb2b148613b443fbc2bfdc30689965
|
|
re2;https://github.com/google/re2/archive/refs/tags/2022-06-01.zip;aa77313b76e91b531ee7f3e45f004c6a502a5374
|
|
safeint;https://github.com/dcleblanc/SafeInt/archive/refs/tags/3.0.28.zip;23f252040ff6cb9f1fd18575b32fa8fb5928daac
|
|
tensorboard;https://github.com/tensorflow/tensorboard/archive/373eb09e4c5d2b3cc2493f0949dc4be6b6a45e81.zip;67b833913605a4f3f499894ab11528a702c2b381
|
|
cutlass;https://github.com/NVIDIA/cutlass/archive/refs/tags/v3.1.0.zip;757f90a795034a89d4f48a79d1f009f7a04c8dee
|
|
utf8_range;https://github.com/protocolbuffers/utf8_range/archive/72c943dea2b9240cd09efde15191e144bc7c7d38.zip;9925739c9debc0efa2adcb194d371a35b6a03156
|
|
extensions;https://github.com/microsoft/onnxruntime-extensions/archive/94142d8391c9791ec71c38336436319a2d4ac7a0.zip;4365ac5140338b4cb75a39944a4be276e3829b3c
|
|
composable_kernel;https://github.com/ROCmSoftwarePlatform/composable_kernel/archive/5356c4a943a35e74d7cdc69486afcb8703b9a59a.zip;522382c2af437e09124287e5879ab64af5b2e299 |