Fix release wheel CUDA index calculation

[fallintoplace]

Summary

• compute the PyTorch CUDA wheel index with a single Python expression
• keep the existing Torch/CUDA version mapping unchanged
• avoid the invalid Python line continuation in the release wheel Dockerfile

Validation

• checked the shell expression for Torch 2.5-2.9 with CUDA 11.8 and 12.8
• ran git diff --check -- .github/actions/build-pytorch-wheel/Dockerfile

[greptile-apps]

Greptile Summary

This PR fixes a Python SyntaxError in the PyTorch wheel install step caused by a backslash-space (\ ) line continuation (the trailing space after \ made it invalid Python syntax). The fix consolidates the multi-line python -c expression into a single semicolon-separated one-liner and also adds the missing trailing newline to the file.

• The Torch/CUDA version mapping (minv/maxv pairs per Torch minor version) is preserved exactly: 2.5–2.7 use 118 as minv, 2.8–2.9 use 126, and maxv values range from 124 to 130.
• The CUDA boundary check (< 120) correctly separates CUDA 11.x (118, 119) from CUDA 12.x (120+), consistent with the original logic.

Confidence Score: 5/5

Safe to merge — the change is a targeted fix to a broken Python expression inside a Dockerfile RUN step, with no functional logic changes.

The old multi-line python -c string had a \ (backslash followed by a space before the newline) on its first continuation line, which is invalid Python and would have caused a SyntaxError at build time. The new single-line form removes that syntax issue, the version lookup table values are bit-for-bit identical to the originals, and the CUDA boundary comparison is unchanged. The file also gains a proper trailing newline. No logic is altered.

No files require special attention.

Important Files Changed

Filename	Overview
.github/actions/build-pytorch-wheel/Dockerfile	Collapses the multi-line Python -c expression into a single line, fixing the invalid backslash-space line continuation that would cause a Python SyntaxError; preserves the Torch/CUDA version mapping verbatim and adds a missing trailing newline.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Docker RUN step] --> B[Extract MATRIX_CUDA_VERSION via awk]
    B --> C[Extract MATRIX_TORCH_VERSION via awk]
    C --> D[Python one-liner: lookup versions dict]
    D --> E{MATRIX_CUDA_VERSION less than 120?}
    E -- Yes: CUDA 11.x --> F[Use minv e.g. cu118 or cu126]
    E -- No: CUDA 12.x --> G[Use maxv e.g. cu124 to cu130]
    F --> H[pip install torch with cu index URL]
    G --> H

Loading

_{Reviews (1): Last reviewed commit: "Fix release wheel CUDA index calculation" | Re-trigger Greptile}