Skip to content

Fix legacy PyTorch compilation and Windows Inductor setup - #51

Merged
OpenUSA merged 1 commit into
mainfrom
fix/torch-compile-portability
Nov 13, 2026
Merged

OpenUSA merged 1 commit into
mainfrom
fix/torch-compile-portability

Conversation

@OpenUSA

@OpenUSA OpenUSA commented Nov 13, 2026 •

Copy link
Copy Markdown
Contributor

Legacy PyTorch full-graph capture failed when expression validation called Tensor.is_complex() or reflected over tensor dtype/shape. Replace these with static dtype and broadcast guards, including Pauli readout bindings, while preserving validation and native derivatives.

Translate observed InvalidCxxCompiler failures into actionable ResourceError messages; retain unrelated backend failures and explicit fallback selection. Report C++ executable discovery and activate the installed MSVC environment in Windows CI. Add lazy execution_info() provenance for Aer and OpenQML tensor backends, and distinguish OpenQML implementations from external tensor/vendor kernels. This change does not implement an independent C++/CUDA engine. Package version: 0.25.1.

Validation:

  • GitHub Linux/Python 3.11 full suite: 2,895 passed, 15 hardware skips; lint, formatting, build and distribution checks passed.
  • GitHub Windows: compilation/fusion suite 16 passed; selected integration suite 1,271 passed, 18 skips. The previously failing Inductor readout test passes with MSVC activated.
  • Legacy-frameworks (Torch 2.2.2), compiled-transforms (Torch 2.4.1), current macOS portability, all three core platforms, distributed tensor correctness and public typing CI passed.
  • Local Torch 2.2.2, 2.4.1 and 2.15 compilation regressions passed; a real circuit/readout/autograd also passed with aot_eager on Torch 2.2.2.
  • Local full suite: 2,892 passed, 15 hardware skips, three Gloo process-group initialization failures caused by this environment denying TCP sockets. Those distributed tests pass in GitHub CI.

The Python 3.10/3.12 full jobs and Intel macOS integration job were still running at this update. No physical GPU throughput or independent CUDA-engine claim is made.

@OpenUSA
OpenUSA merged commit 498e8b7 into main Nov 13, 2026
13 of 14 checks passed
@OpenQML-org OpenQML-org deleted a comment from chatgpt-codex-connector Bot Nov 13, 2026
@OpenUSA
OpenUSA deleted the fix/torch-compile-portability branch November 13, 2026 10:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant