torch_poetry_bootstrap

A simple command-line utility that detects your system's NVIDIA CUDA version and recommends the appropriate PyTorch wheel source. It outputs a tool.poetry.source TOML snippet for Poetry users, ensuring correct PyTorch installation with GPU support.

Installation

If you have Rust installed, the CLI can be installed via cargo, like so:

cargo install torch_poetry_bootstrap

📦 Option 2: Download a Precompiled Binary

If you don’t want to install Rust, you can download a standalone binary from the GitHub Releases:

1. Visit the Releases page.

2. Download the binary appropriate for your system:

   • Windows: torch_poetry_bootstrap.exe
   • Linux: torch_poetry_bootstrap
   • macOS: torch_poetry_bootstrap

3. (Linux/macOS only) Make the binary executable:

   chmod +x torch_poetry_bootstrap
   
   
   

Features

• Detects local CUDA version via nvidia-smi
• Resolves the most compatible PyTorch wheel index (e.g., cu121, cu118, etc.)
• Outputs a copy-pasteable TOML snippet for use in your pyproject.toml
• Supports dry-run and TOML print modes
• Logs messages to stdout and optionally to a file

Example Usage

torch_poetry_bootstrap \
  --patch-pyproject examples/pyproject.toml \
  --output examples/patched.toml

torch_poetry_bootstrap v0.1.15
🔍 Running `nvidia-smi` to detect CUDA version...
✅ Detected CUDA version: 12.3
📄 Loading torch source mapping from JSON...
✅ Selected best matching torch source:
🔗 Source: cu121
🌐 URL: https://download.pytorch.org/whl/cu121
Source 'cu121' with URL 'https://download.pytorch.org/whl/cu121' already exists in the pyproject.toml.
✅ Successfully patched `pyproject.toml` with the new source.: examples/patched.toml
Completed in 1.71s