mirror of
https://github.com/zebrajr/pytorch.git
synced 2026-01-15 12:15:51 +00:00
b04f0b1b27
Fixes #171385 Summary Replaced hardcoded `/tmp` paths with `tempfile.gettempdir()` in production code and key test files to allow users to configure temp directory location via environment variables (e.g., `TMPDIR`). Motivation Users need to specify custom temp directory locations when: - `/tmp` has limited storage or quota restrictions - Running on Windows (where `/tmp` doesn't exist) - Using containerized environments with custom temp directories Changes Modified 7 files to use `tempfile.gettempdir()`: - `torch/_inductor/debug.py` - Inductor saved args - `torch/_dynamo/convert_frame.py` - cProfile outputs - `torch/_inductor/compile_fx_ext.py` - Debug serialization files - `torch/distributed/elastic/agent/server/local_elastic_agent.py` - Watchdog timers - `torch/distributed/_symmetric_memory/_nvshmem_triton.py` - NVSHMEM PTX files - `benchmarks/dynamo/torchbench.py` - KALDI_ROOT env var - `test/test_cuda.py` - GDS filesystem tests Scope Note: This PR focuses on production code and critical test files (~7 files). The codebase has ~96 `/tmp` references across 53 files, but many are in: - Comments/docstrings (documentation examples) - Example scripts (functorch examples) - Less critical test files These were intentionally excluded to keep the PR focused and minimize risk of breaking existing workflows. Follow-up PRs can address remaining instances if needed. Backward Compatibility Fully backward compatible - when `TMPDIR` is not set, `tempfile.gettempdir()` defaults to `/tmp` on Unix/Linux systems, maintaining existing behavior. Testing - Verified with `git diff` that only intended changes were made - All modified paths now respect system temp directory configuration - Users can test with: `export TMPDIR=/custom/path` Pull Request resolved: https://github.com/pytorch/pytorch/pull/171446 Approved by: https://github.com/malfet Co-authored-by: Nikita Shulga <2453524+malfet@users.noreply.github.com>
PyTorch Benchmarks
This folder contains scripts that produce reproducible timings of various PyTorch features.
It also provides mechanisms to compare PyTorch with other frameworks.
Setup environment
Make sure you're on a machine with CUDA, torchvision, and pytorch installed. Install in the following order:
# Install torchvision. It comes with the pytorch stable release binary
python -m pip install torch torchvision
# Install the latest pytorch master from source.
# It should supersede the installation from the release binary.
cd $PYTORCH_HOME
python -m pip install --no-build-isolation -v -e .
# Check the pytorch installation version
python -c "import torch; print(torch.__version__)"
Benchmark List
Please refer to each subfolder to discover each benchmark suite. Links are provided where descriptions exist: