BSD Ports

ai/pytorch 2.13.0_1 MidnightBSD

Tensors and dynamic neural networks in Python (C++ library)

System
MidnightBSD
Repository
mports
Maintainer
[email protected]
Last update
Commit
9f1f9b533d
License
bsd3
Categories
ai

Description

PyTorch is a Python package that provides two high-level features: * Tensor computation (like NumPy) with strong GPU acceleration * Deep neural networks built on a tape-based autograd system

In the other systems

SystemPortVersion
MidnightBSDai/pytorch2.13.0_1
FreeBSDmisc/pytorch2.13.0
NetBSDwip/py-torchv2.12.1
GhostBSDmisc/pytorch2.13.0

Install

Binary package:

mport install pytorch

From mports:

cd /usr/mports/ai/pytorch && make install clean

To depend on this port in another port's Makefile:

RUN_DEPENDS=	pytorch>0:ai/pytorch
LIB_DEPENDS=	libc10.so:ai/pytorch

Dependencies

As this port's Makefile names them.

Build 2

Library 9

Build dependencies 11
Run dependencies 9

No port depends on this one.

Configuration

Options name: ai_pytorch

OptionDefaultDescription
DISTRIBUTEDoffDistributed training support (gloo, MPI, tensorpipe)
PYTHONon
USES
compiler:c++17-lang cmake:noninja gmake localbase:ldflags python
Service scripts
none

Files

pkg-plist 9370
  • _C.so
  • _C/_VariableFunctions.pyi
  • _C/__init__.pyi
  • _C/_acc/__init__.pyi
  • _C/_aoti.pyi
  • _C/_autograd.pyi
  • _C/_cpu.pyi
  • _C/_cudnn.pyi
  • _C/_cusparselt.pyi
  • _C/_distributed.pyi
  • _C/_distributed_autograd.pyi
  • _C/_distributed_c10d.pyi
  • _C/_distributed_rpc.pyi
  • _C/_distributed_rpc_testing.pyi
  • _C/_dynamo/__init__.pyi
  • _C/_dynamo/compiled_autograd.pyi
  • _C/_dynamo/eval_frame.pyi
  • _C/_dynamo/guards.pyi
  • _C/_export/__init__.pyi
  • _C/_export/pt2_archive_constants.pyi
  • _C/_functionalization.pyi
  • _C/_functions.pyi
  • _C/_functorch.pyi
  • _C/_instruction_counter.pyi
  • _C/_itt.pyi
  • _C/_jit_tree_views.pyi
  • _C/_lazy.pyi
  • _C/_lazy_ts_backend.pyi
  • _C/_monitor.pyi
  • _C/_nn.pyi
  • _C/_nvtx.pyi
  • _C/_onnx.pyi
  • _C/_profiler.pyi
  • _C/_verbose.pyi
  • _C_flatbuffer/__init__.pyi
  • _VF.pyi
  • _dynamo/graph_break_registry.json
  • _export/serde/export_schema.thrift
  • _export/serde/schema.yaml
  • _inductor/codegen/aoti_runtime/interface.cpp
  • _inductor/kernel/flex/templates/common.py.jinja
  • _inductor/kernel/flex/templates/flash_attention.py.jinja
  • _inductor/kernel/flex/templates/flash_attention_backward.py.jinja
  • _inductor/kernel/flex/templates/flex_attention.py.jinja
  • _inductor/kernel/flex/templates/flex_backwards.py.jinja
  • _inductor/kernel/flex/templates/flex_decode.py.jinja
  • _inductor/kernel/flex/templates/utilities.py.jinja
  • _inductor/kernel/templates/cutedsl_mm_grouped.py.jinja
  • _inductor/kernel/templates/nv_universal_gemm.py.jinja
  • _inductor/kernel/templates/triton_blackwell_ws_persistent_device_tma_mm.py.jinja
  • _inductor/kernel/templates/triton_bmm.py.jinja
  • _inductor/kernel/templates/triton_conv2d_bwd_input.py.jinja
  • _inductor/kernel/templates/triton_conv2d_bwd_weight.py.jinja
  • _inductor/kernel/templates/triton_depthwise_conv.py.jinja
  • _inductor/kernel/templates/triton_epilogue_scaled_mm.py.jinja
  • _inductor/kernel/templates/triton_main_loop_scaled_mm.py.jinja
  • _inductor/kernel/templates/triton_mm.py.jinja
  • _inductor/kernel/templates/triton_mm_grouped.py.jinja
  • _inductor/kernel/templates/triton_mm_rocm.py.jinja
  • _inductor/kernel/templates/triton_persistent_mm.py.jinja
  • _inductor/kernel/templates/triton_persistent_tma_mm.py.jinja
  • _inductor/script.ld
  • csrc/inductor/aoti_runtime/model.h
  • distributed/optim/zero_redundancy_optimizer.pyi
  • jit/_script.pyi
  • nn/functional.pyi
  • nn/parameter.pyi
  • optim/_multi_tensor/__init__.pyi
  • py.typed
  • return_types.pyi
  • testing/_internal/generated/annotated_fn_args.py
  • utils/_config_typing.pyi
  • utils/benchmark/utils/timeit_template.cpp
  • utils/benchmark/utils/valgrind_wrapper/compat_bindings.cpp
  • utils/benchmark/utils/valgrind_wrapper/timer_callgrind_template.cpp
  • utils/data/datapipes/datapipe.pyi
  • utils/model_dump/code.js
  • utils/model_dump/htm.mjs
  • utils/model_dump/preact.mjs
  • utils/model_dump/skeleton.html
  • version.py
  • bin/torch_shm_manager
  • include/ATen/ATen.h
  • include/ATen/AccumulateType.h
  • include/ATen/ArrayRef.h
  • include/ATen/Backend.h
  • include/ATen/Backtrace.h
  • include/ATen/BlasBackend.h
  • include/ATen/CPUApplyUtils.h
  • include/ATen/CPUFunctions.h
  • include/ATen/CPUFunctions_inl.h
  • include/ATen/CPUGeneratorImpl.h
  • include/ATen/CUDAFunctions.h
  • include/ATen/CUDAFunctions_inl.h
  • include/ATen/CachedTensorUtils.h
  • include/ATen/CollapseDims.h
  • include/ATen/CompositeExplicitAutogradFunctions.h
  • include/ATen/CompositeExplicitAutogradFunctions_inl.h
  • include/ATen/CompositeExplicitAutogradNonFunctionalFunctions.h
  • include/ATen/CompositeExplicitAutogradNonFunctionalFunctions_inl.h

The first 100 are listed.

distinfo 1
FileSizeSHA256
pytorch/pytorch-v2.13.0.tar.gz4.85291648e+0866614a19060f69cfd63cd0295f65a1241bd15df2fa65c60ae51066c11c2ce812
Master sites 1
  • https://github.com/pytorch/pytorch/releases/download/v2.13.0/

Commit history 3

DateCommitCreditsLog message
9f1f9b533d Lucas Holt ai/pytorch: add optional distributed training support (#852)
Adds a `DISTRIBUTED` option, **off by default**, that enables `USE_DISTRIBUTED` and builds the gloo, MPI and tensorpipe backends. Follow-up to #851, which asked whether the libuv patch it carried was dead weight. It isn't — this is what exercises it. ## Why it was off Upstream disables `USE_DISTRIBUTED` on every platform that isn't Linux or Windows (`CMakeLists.txt:200` — *"not tested and likely won't work without additional changes"*), and the FreeBSD port leaves it that way. Two things were needed to make it build here. ### 1. libuv (patch already in tree) `CMAKE_SYSTEM_NAME` is `"MidnightBSD"`, so libuv's `MATCHES "FreeBSD"` guards never fire and it gets **no BSD sources at all**. The patch from #851 fixes that; with it, the build compiles `kqueue.c`, `freebsd.c`, …
58a7cfb2ad Lucas Holt ai/pytorch: update to 2.13.0 (#851)
dee1e47e52 Lucas Holt ai: add new category and move AI/ML ports from misc
Create the ai category (Artificial intelligence and machine learning) and relocate 42 AI/ML ports from misc into it. - Register ai in Mk/bsd.mport.mk VALID_CATEGORIES and the top-level Makefile - Add ai/Makefile; drop the moved entries from misc/Makefile (also removes the stale linux-claude SUBDIR left over from the claude-code rename) - git mv 42 ports and set their primary CATEGORIES to ai - Rewrite misc/<port> dependency origins to ai/<port> in external dependents (devel/py-beartype, math/py-faiss, multimedia/py-torchcodec, textproc/py-tokenizers) and among the moved ports themselves - Record the origin migrations in MOVED Ports moved: chatgpt-shell-cli, claude-code, claude-code-legacy, clawhub, codex, gemini-cli, github-copilot-cli, grok-cli, lean-ctx, llama-cpp, ollama, …