| commit | 603bde1de376cd242bf365ef361a281809c8e6ab | [log] [tgz] |
|---|---|---|
| author | Scott Wolchok <swolchok@meta.com> | Wed May 29 17:11:08 2024 -0700 |
| committer | PyTorch MergeBot <pytorchmergebot@users.noreply.github.com> | Thu May 30 23:28:17 2024 +0000 |
| tree | c34f4ab91ae1f4f18091ff3c1cbcd4a85a904a3e | |
| parent | 74b89b9283373b2137fcef74508dc9a38c8097c9 [diff] |
Use efficient ARM fp16 dot product for gemm_transa_ general case (#127451) Summary: This doesn't change the overall gemm algorithm away from repeated dot products, just uses our efficient fp16 dot product developed for the gemv case. It seems to improve performance for every prompt length I tested. Test Plan: Use https://github.com/malfet/llm_experiments/blob/main/benchmarks/benchmark_torch_mm.py , edited to test only the trans_b (really gemm_transa_) case for the sizes outlined in the output. Before: ``` Matrix-vector: m=8, n=128, k=1 ==================== trans_b torch.float32 1.05 usec trans_b torch.float16 0.97 usec trans_b torch.bfloat16 1.06 usec m=128, n=8, k=1 ==================== trans_b torch.float32 0.80 usec trans_b torch.float16 0.97 usec trans_b torch.bfloat16 1.00 usec m=4096, n=4096, k=1 ==================== trans_b torch.float32 2160.75 usec trans_b torch.float16 659.77 usec trans_b torch.bfloat16 3800.13 usec m=11008, n=4096, k=1 ==================== trans_b torch.float32 6343.68 usec trans_b torch.float16 1789.42 usec trans_b torch.bfloat16 10098.34 usec m=4096, n=11008, k=1 ==================== trans_b torch.float32 6217.20 usec trans_b torch.float16 1874.47 usec trans_b torch.bfloat16 10490.30 usec m=32000, n=4096, k=1 ==================== trans_b torch.float32 17934.45 usec trans_b torch.float16 5323.81 usec trans_b torch.bfloat16 29320.80 usec Matrix-matrix (prompt len 4: m=8, n=128, k=4 ==================== trans_b torch.float32 2.40 usec trans_b torch.float16 1.22 usec trans_b torch.bfloat16 1.22 usec m=128, n=8, k=4 ==================== trans_b torch.float32 1.52 usec trans_b torch.float16 1.33 usec trans_b torch.bfloat16 1.77 usec m=4096, n=4096, k=4 ==================== trans_b torch.float32 4317.09 usec trans_b torch.float16 15541.04 usec trans_b torch.bfloat16 15032.29 usec m=11008, n=4096, k=4 ==================== trans_b torch.float32 6191.19 usec trans_b torch.float16 40436.29 usec trans_b torch.bfloat16 40626.93 usec m=4096, n=11008, k=4 ==================== trans_b torch.float32 6049.22 usec trans_b torch.float16 42367.16 usec trans_b torch.bfloat16 42482.43 usec m=32000, n=4096, k=4 ==================== trans_b torch.float32 17611.36 usec trans_b torch.float16 117368.54 usec trans_b torch.bfloat16 116958.85 usec Matrix-matrix (prompt len 8: m=8, n=128, k=8 ==================== trans_b torch.float32 1.04 usec trans_b torch.float16 1.71 usec trans_b torch.bfloat16 1.74 usec m=128, n=8, k=8 ==================== trans_b torch.float32 2.10 usec trans_b torch.float16 2.01 usec trans_b torch.bfloat16 2.91 usec m=4096, n=4096, k=8 ==================== trans_b torch.float32 2456.23 usec trans_b torch.float16 30112.76 usec trans_b torch.bfloat16 29941.58 usec m=11008, n=4096, k=8 ==================== trans_b torch.float32 6236.12 usec trans_b torch.float16 80361.22 usec trans_b torch.bfloat16 80466.64 usec m=4096, n=11008, k=8 ==================== trans_b torch.float32 6236.10 usec trans_b torch.float16 82990.74 usec trans_b torch.bfloat16 83899.80 usec m=32000, n=4096, k=8 ==================== trans_b torch.float32 17606.43 usec trans_b torch.float16 234397.38 usec trans_b torch.bfloat16 237057.29 usec Matrix-matrix (prompt len 16: m=8, n=128, k=16 ==================== trans_b torch.float32 1.31 usec trans_b torch.float16 2.67 usec trans_b torch.bfloat16 2.72 usec m=128, n=8, k=16 ==================== trans_b torch.float32 1.66 usec trans_b torch.float16 3.36 usec trans_b torch.bfloat16 5.18 usec m=4096, n=4096, k=16 ==================== trans_b torch.float32 2504.24 usec trans_b torch.float16 60896.53 usec trans_b torch.bfloat16 59852.49 usec m=11008, n=4096, k=16 ==================== trans_b torch.float32 6407.11 usec trans_b torch.float16 163294.92 usec trans_b torch.bfloat16 161199.10 usec m=4096, n=11008, k=16 ==================== trans_b torch.float32 6132.30 usec trans_b torch.float16 167244.77 usec trans_b torch.bfloat16 170064.35 usec m=32000, n=4096, k=16 ==================== trans_b torch.float32 17635.56 usec trans_b torch.float16 475020.00 usec trans_b torch.bfloat16 476332.29 usec Matrix-matrix (prompt len 32: m=8, n=128, k=32 ==================== trans_b torch.float32 1.40 usec trans_b torch.float16 4.67 usec trans_b torch.bfloat16 4.80 usec m=128, n=8, k=32 ==================== trans_b torch.float32 1.24 usec trans_b torch.float16 6.10 usec trans_b torch.bfloat16 10.03 usec m=4096, n=4096, k=32 ==================== trans_b torch.float32 2660.63 usec trans_b torch.float16 122436.04 usec trans_b torch.bfloat16 121687.96 usec m=11008, n=4096, k=32 ==================== trans_b torch.float32 6405.60 usec trans_b torch.float16 324708.42 usec trans_b torch.bfloat16 324866.67 usec m=4096, n=11008, k=32 ==================== trans_b torch.float32 6566.74 usec trans_b torch.float16 330801.04 usec trans_b torch.bfloat16 332561.79 usec m=32000, n=4096, k=32 ==================== trans_b torch.float32 18610.84 usec trans_b torch.float16 944578.75 usec trans_b torch.bfloat16 940674.33 usec Matrix-matrix (prompt len 128: m=8, n=128, k=128 ==================== trans_b torch.float32 2.48 usec trans_b torch.float16 16.43 usec trans_b torch.bfloat16 17.11 usec m=128, n=8, k=128 ==================== trans_b torch.float32 1.83 usec trans_b torch.float16 22.31 usec trans_b torch.bfloat16 37.00 usec m=4096, n=4096, k=128 ==================== trans_b torch.float32 4806.59 usec trans_b torch.float16 485338.83 usec trans_b torch.bfloat16 478835.08 usec m=11008, n=4096, k=128 ==================== trans_b torch.float32 12109.51 usec trans_b torch.float16 1300928.58 usec trans_b torch.bfloat16 1293181.63 usec m=4096, n=11008, k=128 ==================== trans_b torch.float32 11223.70 usec trans_b torch.float16 1326119.92 usec trans_b torch.bfloat16 1330395.12 usec m=32000, n=4096, k=128 ==================== trans_b torch.float32 33485.34 usec trans_b torch.float16 3869227.17 usec trans_b torch.bfloat16 3792905.00 usec ``` After: ``` Matrix-vector: m=8, n=128, k=1 ==================== trans_b torch.float32 0.75 usec trans_b torch.float16 0.71 usec trans_b torch.bfloat16 0.81 usec m=128, n=8, k=1 ==================== trans_b torch.float32 0.75 usec trans_b torch.float16 0.93 usec trans_b torch.bfloat16 0.98 usec m=4096, n=4096, k=1 ==================== trans_b torch.float32 2194.31 usec trans_b torch.float16 661.27 usec trans_b torch.bfloat16 3758.42 usec m=11008, n=4096, k=1 ==================== trans_b torch.float32 5792.04 usec trans_b torch.float16 1789.98 usec trans_b torch.bfloat16 10120.67 usec m=4096, n=11008, k=1 ==================== trans_b torch.float32 6101.22 usec trans_b torch.float16 1927.34 usec trans_b torch.bfloat16 10469.47 usec m=32000, n=4096, k=1 ==================== trans_b torch.float32 18353.20 usec trans_b torch.float16 5161.06 usec trans_b torch.bfloat16 29601.69 usec Matrix-matrix (prompt len 4: m=8, n=128, k=4 ==================== trans_b torch.float32 2.14 usec trans_b torch.float16 0.85 usec trans_b torch.bfloat16 1.19 usec m=128, n=8, k=4 ==================== trans_b torch.float32 1.47 usec trans_b torch.float16 1.85 usec trans_b torch.bfloat16 1.75 usec m=4096, n=4096, k=4 ==================== trans_b torch.float32 4416.40 usec trans_b torch.float16 2688.36 usec trans_b torch.bfloat16 14987.33 usec m=11008, n=4096, k=4 ==================== trans_b torch.float32 6140.24 usec trans_b torch.float16 7467.26 usec trans_b torch.bfloat16 40295.52 usec m=4096, n=11008, k=4 ==================== trans_b torch.float32 6143.10 usec trans_b torch.float16 7298.04 usec trans_b torch.bfloat16 41393.43 usec m=32000, n=4096, k=4 ==================== trans_b torch.float32 17650.72 usec trans_b torch.float16 21346.63 usec trans_b torch.bfloat16 116849.98 usec Matrix-matrix (prompt len 8: m=8, n=128, k=8 ==================== trans_b torch.float32 1.05 usec trans_b torch.float16 1.03 usec trans_b torch.bfloat16 1.69 usec m=128, n=8, k=8 ==================== trans_b torch.float32 2.05 usec trans_b torch.float16 3.08 usec trans_b torch.bfloat16 2.95 usec m=4096, n=4096, k=8 ==================== trans_b torch.float32 2323.99 usec trans_b torch.float16 5265.45 usec trans_b torch.bfloat16 29942.40 usec m=11008, n=4096, k=8 ==================== trans_b torch.float32 6202.01 usec trans_b torch.float16 14677.90 usec trans_b torch.bfloat16 80625.18 usec m=4096, n=11008, k=8 ==================== trans_b torch.float32 6112.05 usec trans_b torch.float16 14340.52 usec trans_b torch.bfloat16 82799.99 usec m=32000, n=4096, k=8 ==================== trans_b torch.float32 17650.65 usec trans_b torch.float16 42551.43 usec trans_b torch.bfloat16 236081.08 usec Matrix-matrix (prompt len 16: m=8, n=128, k=16 ==================== trans_b torch.float32 1.26 usec trans_b torch.float16 1.34 usec trans_b torch.bfloat16 2.69 usec m=128, n=8, k=16 ==================== trans_b torch.float32 1.60 usec trans_b torch.float16 5.81 usec trans_b torch.bfloat16 5.34 usec m=4096, n=4096, k=16 ==================== trans_b torch.float32 2328.05 usec trans_b torch.float16 10526.58 usec trans_b torch.bfloat16 60028.28 usec m=11008, n=4096, k=16 ==================== trans_b torch.float32 6243.35 usec trans_b torch.float16 28505.08 usec trans_b torch.bfloat16 163670.15 usec m=4096, n=11008, k=16 ==================== trans_b torch.float32 5870.11 usec trans_b torch.float16 28597.89 usec trans_b torch.bfloat16 165404.88 usec m=32000, n=4096, k=16 ==================== trans_b torch.float32 17746.27 usec trans_b torch.float16 83393.87 usec trans_b torch.bfloat16 472313.13 usec Matrix-matrix (prompt len 32: m=8, n=128, k=32 ==================== trans_b torch.float32 1.35 usec trans_b torch.float16 2.01 usec trans_b torch.bfloat16 4.68 usec m=128, n=8, k=32 ==================== trans_b torch.float32 1.19 usec trans_b torch.float16 10.98 usec trans_b torch.bfloat16 10.13 usec m=4096, n=4096, k=32 ==================== trans_b torch.float32 2525.29 usec trans_b torch.float16 23106.71 usec trans_b torch.bfloat16 122987.04 usec m=11008, n=4096, k=32 ==================== trans_b torch.float32 6131.34 usec trans_b torch.float16 57537.41 usec trans_b torch.bfloat16 327825.00 usec m=4096, n=11008, k=32 ==================== trans_b torch.float32 6395.01 usec trans_b torch.float16 57456.33 usec trans_b torch.bfloat16 331325.58 usec m=32000, n=4096, k=32 ==================== trans_b torch.float32 19078.68 usec trans_b torch.float16 167735.08 usec trans_b torch.bfloat16 975736.88 usec Matrix-matrix (prompt len 128: m=8, n=128, k=128 ==================== trans_b torch.float32 2.40 usec trans_b torch.float16 6.07 usec trans_b torch.bfloat16 16.83 usec m=128, n=8, k=128 ==================== trans_b torch.float32 1.78 usec trans_b torch.float16 40.35 usec trans_b torch.bfloat16 37.21 usec m=4096, n=4096, k=128 ==================== trans_b torch.float32 4827.60 usec trans_b torch.float16 84341.24 usec trans_b torch.bfloat16 478917.75 usec m=11008, n=4096, k=128 ==================== trans_b torch.float32 11879.96 usec trans_b torch.float16 226484.33 usec trans_b torch.bfloat16 1289465.50 usec m=4096, n=11008, k=128 ==================== trans_b torch.float32 10707.75 usec trans_b torch.float16 229200.58 usec trans_b torch.bfloat16 1327416.67 usec m=32000, n=4096, k=128 ==================== trans_b torch.float32 33306.32 usec trans_b torch.float16 662898.21 usec trans_b torch.bfloat16 3815866.63 usec ``` torch.float16 performance seems to be improved for all except the m=128, n=8, k=128 case, where it is roughly neutral. This case motivated the addition of the "first-tier tail fixup" in the dot kernel. Pull Request resolved: https://github.com/pytorch/pytorch/pull/127451 Approved by: https://github.com/malfet ghstack dependencies: #127435
PyTorch is a Python package that provides two high-level features:
You can reuse your favorite Python packages such as NumPy, SciPy, and Cython to extend PyTorch when needed.
Our trunk health (Continuous Integration signals) can be found at hud.pytorch.org.
At a granular level, PyTorch is a library that consists of the following components:
| Component | Description |
|---|---|
| torch | A Tensor library like NumPy, with strong GPU support |
| torch.autograd | A tape-based automatic differentiation library that supports all differentiable Tensor operations in torch |
| torch.jit | A compilation stack (TorchScript) to create serializable and optimizable models from PyTorch code |
| torch.nn | A neural networks library deeply integrated with autograd designed for maximum flexibility |
| torch.multiprocessing | Python multiprocessing, but with magical memory sharing of torch Tensors across processes. Useful for data loading and Hogwild training |
| torch.utils | DataLoader and other utility functions for convenience |
Usually, PyTorch is used either as:
Elaborating Further:
If you use NumPy, then you have used Tensors (a.k.a. ndarray).
PyTorch provides Tensors that can live either on the CPU or the GPU and accelerates the computation by a huge amount.
We provide a wide variety of tensor routines to accelerate and fit your scientific computation needs such as slicing, indexing, mathematical operations, linear algebra, reductions. And they are fast!
PyTorch has a unique way of building neural networks: using and replaying a tape recorder.
Most frameworks such as TensorFlow, Theano, Caffe, and CNTK have a static view of the world. One has to build a neural network and reuse the same structure again and again. Changing the way the network behaves means that one has to start from scratch.
With PyTorch, we use a technique called reverse-mode auto-differentiation, which allows you to change the way your network behaves arbitrarily with zero lag or overhead. Our inspiration comes from several research papers on this topic, as well as current and past work such as torch-autograd, autograd, Chainer, etc.
While this technique is not unique to PyTorch, it's one of the fastest implementations of it to date. You get the best of speed and flexibility for your crazy research.
PyTorch is not a Python binding into a monolithic C++ framework. It is built to be deeply integrated into Python. You can use it naturally like you would use NumPy / SciPy / scikit-learn etc. You can write your new neural network layers in Python itself, using your favorite libraries and use packages such as Cython and Numba. Our goal is to not reinvent the wheel where appropriate.
PyTorch is designed to be intuitive, linear in thought, and easy to use. When you execute a line of code, it gets executed. There isn't an asynchronous view of the world. When you drop into a debugger or receive error messages and stack traces, understanding them is straightforward. The stack trace points to exactly where your code was defined. We hope you never spend hours debugging your code because of bad stack traces or asynchronous and opaque execution engines.
PyTorch has minimal framework overhead. We integrate acceleration libraries such as Intel MKL and NVIDIA (cuDNN, NCCL) to maximize speed. At the core, its CPU and GPU Tensor and neural network backends are mature and have been tested for years.
Hence, PyTorch is quite fast — whether you run small or large neural networks.
The memory usage in PyTorch is extremely efficient compared to Torch or some of the alternatives. We've written custom memory allocators for the GPU to make sure that your deep learning models are maximally memory efficient. This enables you to train bigger deep learning models than before.
Writing new neural network modules, or interfacing with PyTorch's Tensor API was designed to be straightforward and with minimal abstractions.
You can write new neural network layers in Python using the torch API or your favorite NumPy-based libraries such as SciPy.
If you want to write your layers in C/C++, we provide a convenient extension API that is efficient and with minimal boilerplate. No wrapper code needs to be written. You can see a tutorial here and an example here.
Commands to install binaries via Conda or pip wheels are on our website: https://pytorch.org/get-started/locally/
Python wheels for NVIDIA's Jetson Nano, Jetson TX1/TX2, Jetson Xavier NX/AGX, and Jetson AGX Orin are provided here and the L4T container is published here
They require JetPack 4.2 and above, and @dusty-nv and @ptrblck are maintaining them.
If you are installing from source, you will need:
We highly recommend installing an Anaconda environment. You will get a high-quality BLAS library (MKL) and you get controlled dependency versions regardless of your Linux distro.
If you want to compile with CUDA support, select a supported version of CUDA from our support matrix, then install the following:
Note: You could refer to the cuDNN Support Matrix for cuDNN versions with the various supported CUDA, CUDA driver and NVIDIA hardware
If you want to disable CUDA support, export the environment variable USE_CUDA=0. Other potentially useful environment variables may be found in setup.py.
If you are building for NVIDIA's Jetson platforms (Jetson Nano, TX1, TX2, AGX Xavier), Instructions to install PyTorch for Jetson Nano are available here
If you want to compile with ROCm support, install
If you want to disable ROCm support, export the environment variable USE_ROCM=0. Other potentially useful environment variables may be found in setup.py.
If you want to compile with Intel GPU support, follow these
If you want to disable Intel GPU support, export the environment variable USE_XPU=0. Other potentially useful environment variables may be found in setup.py.
Common
conda install cmake ninja # Run this command from the PyTorch directory after cloning the source code using the “Get the PyTorch Source“ section below pip install -r requirements.txt
On Linux
conda install intel::mkl-static intel::mkl-include # CUDA only: Add LAPACK support for the GPU if needed conda install -c pytorch magma-cuda121 # or the magma-cuda* that matches your CUDA version from https://anaconda.org/pytorch/repo # (optional) If using torch.compile with inductor/triton, install the matching version of triton # Run from the pytorch directory after cloning make triton
On MacOS
# Add this package on intel x86 processor machines only conda install intel::mkl-static intel::mkl-include # Add these packages if torch.distributed is needed conda install pkg-config libuv
On Windows
conda install intel::mkl-static intel::mkl-include # Add these packages if torch.distributed is needed. # Distributed package support on Windows is a prototype feature and is subject to changes. conda install -c conda-forge libuv=1.39
git clone --recursive https://github.com/pytorch/pytorch cd pytorch # if you are updating an existing checkout git submodule sync git submodule update --init --recursive
On Linux
If you would like to compile PyTorch with new C++ ABI enabled, then first run this command:
export _GLIBCXX_USE_CXX11_ABI=1
If you're compiling for AMD ROCm then first run this command:
# Only run this if you're compiling for ROCm python tools/amd_build/build_amd.py
Install PyTorch
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"} python setup.py develop
Aside: If you are using Anaconda, you may experience an error caused by the linker:
build/temp.linux-x86_64-3.7/torch/csrc/stub.o: file not recognized: file format not recognized collect2: error: ld returned 1 exit status error: command 'g++' failed with exit status 1This is caused by
ldfrom the Conda environment shadowing the systemld. You should use a newer version of Python that fixes this issue. The recommended Python version is 3.8.1+.
On macOS
python3 setup.py develop
On Windows
Choose Correct Visual Studio Version.
PyTorch CI uses Visual C++ BuildTools, which come with Visual Studio Enterprise, Professional, or Community Editions. You can also install the build tools from https://visualstudio.microsoft.com/visual-cpp-build-tools/. The build tools do not come with Visual Studio Code by default.
If you want to build legacy python code, please refer to Building on legacy code and CUDA
CPU-only builds
In this mode PyTorch computations will run on your CPU, not your GPU
conda activate python setup.py develop
Note on OpenMP: The desired OpenMP implementation is Intel OpenMP (iomp). In order to link against iomp, you'll need to manually download the library and set up the building environment by tweaking CMAKE_INCLUDE_PATH and LIB. The instruction here is an example for setting up both MKL and Intel OpenMP. Without these configurations for CMake, Microsoft Visual C OpenMP runtime (vcomp) will be used.
CUDA based build
In this mode PyTorch computations will leverage your GPU via CUDA for faster number crunching
NVTX is needed to build Pytorch with CUDA. NVTX is a part of CUDA distributive, where it is called “Nsight Compute”. To install it onto an already installed CUDA run CUDA installation once again and check the corresponding checkbox. Make sure that CUDA with Nsight Compute is installed after Visual Studio.
Currently, VS 2017 / 2019, and Ninja are supported as the generator of CMake. If ninja.exe is detected in PATH, then Ninja will be used as the default generator, otherwise, it will use VS 2017 / 2019.
If Ninja is selected as the generator, the latest MSVC will get selected as the underlying toolchain.
Additional libraries such as Magma, oneDNN, a.k.a. MKLDNN or DNNL, and Sccache are often needed. Please refer to the installation-helper to install them.
You can refer to the build_pytorch.bat script for some other environment variables configurations
cmd :: Set the environment variables after you have downloaded and unzipped the mkl package, :: else CMake would throw an error as `Could NOT find OpenMP`. set CMAKE_INCLUDE_PATH={Your directory}\mkl\include set LIB={Your directory}\mkl\lib;%LIB% :: Read the content in the previous section carefully before you proceed. :: [Optional] If you want to override the underlying toolset used by Ninja and Visual Studio with CUDA, please run the following script block. :: "Visual Studio 2019 Developer Command Prompt" will be run automatically. :: Make sure you have CMake >= 3.12 before you do this when you use the Visual Studio generator. set CMAKE_GENERATOR_TOOLSET_VERSION=14.27 set DISTUTILS_USE_SDK=1 for /f "usebackq tokens=*" %i in (`"%ProgramFiles(x86)%\Microsoft Visual Studio\Installer\vswhere.exe" -version [15^,17^) -products * -latest -property installationPath`) do call "%i\VC\Auxiliary\Build\vcvarsall.bat" x64 -vcvars_ver=%CMAKE_GENERATOR_TOOLSET_VERSION% :: [Optional] If you want to override the CUDA host compiler set CUDAHOSTCXX=C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\VC\Tools\MSVC\14.27.29110\bin\HostX64\x64\cl.exe python setup.py develop
You can adjust the configuration of cmake variables optionally (without building first), by doing the following. For example, adjusting the pre-detected directories for CuDNN or BLAS can be done with such a step.
On Linux
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"} python setup.py build --cmake-only ccmake build # or cmake-gui build
On macOS
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"} MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ python setup.py build --cmake-only ccmake build # or cmake-gui build
You can also pull a pre-built docker image from Docker Hub and run with docker v19.03+
docker run --gpus all --rm -ti --ipc=host pytorch/pytorch:latest
Please note that PyTorch uses shared memory to share data between processes, so if torch multiprocessing is used (e.g. for multithreaded data loaders) the default shared memory segment size that container runs with is not enough, and you should increase shared memory size either with --ipc=host or --shm-size command line options to nvidia-docker run.
NOTE: Must be built with a docker version > 18.06
The Dockerfile is supplied to build images with CUDA 11.1 support and cuDNN v8. You can pass PYTHON_VERSION=x.y make variable to specify which Python version is to be used by Miniconda, or leave it unset to use the default.
make -f docker.Makefile # images are tagged as docker.io/${your_docker_username}/pytorch
You can also pass the CMAKE_VARS="..." environment variable to specify additional CMake variables to be passed to CMake during the build. See setup.py for the list of available variables.
make -f docker.Makefile
To build documentation in various formats, you will need Sphinx and the readthedocs theme.
cd docs/ pip install -r requirements.txt
You can then build the documentation by running make <format> from the docs/ folder. Run make to get a list of all available output formats.
If you get a katex error run npm install katex. If it persists, try npm install -g katex
Note: if you installed
nodejswith a different package manager (e.g.,conda) thennpmwill probably install a version ofkatexthat is not compatible with your version ofnodejsand doc builds will fail. A combination of versions that is known to work isnode@6.13.1andkatex@0.13.18. To install the latter withnpmyou can runnpm install -g katex@0.13.18
Installation instructions and binaries for previous PyTorch versions may be found on our website.
Three-pointers to get you started:
Typically, PyTorch has three minor releases a year. Please let us know if you encounter a bug by filing an issue.
We appreciate all contributions. If you are planning to contribute back bug-fixes, please do so without any further discussion.
If you plan to contribute new features, utility functions, or extensions to the core, please first open an issue and discuss the feature with us. Sending a PR without discussion might end up resulting in a rejected PR because we might be taking the core in a different direction than you might be aware of.
To learn more about making a contribution to Pytorch, please see our Contribution page. For more information about PyTorch releases, see Release page.
PyTorch is a community-driven project with several skillful engineers and researchers contributing to it.
PyTorch is currently maintained by Soumith Chintala, Gregory Chanan, Dmytro Dzhulgakov, Edward Yang, and Nikita Shulga with major contributions coming from hundreds of talented individuals in various forms and means. A non-exhaustive but growing list needs to mention: Trevor Killeen, Sasank Chilamkurthy, Sergey Zagoruyko, Adam Lerer, Francisco Massa, Alykhan Tejani, Luca Antiga, Alban Desmaison, Andreas Koepf, James Bradbury, Zeming Lin, Yuandong Tian, Guillaume Lample, Marat Dukhan, Natalia Gimelshein, Christian Sarofeen, Martin Raison, Edward Yang, Zachary Devito.
Note: This project is unrelated to hughperkins/pytorch with the same name. Hugh is a valuable contributor to the Torch community and has helped with many things Torch and PyTorch.
PyTorch has a BSD-style license, as found in the LICENSE file.