Overview

Fixed broken CLI API and improved JIT function @torch.compile(...) decorator usage. Also greatly improved README installation guide for CUDA compatible installlation.

Changes

• Explictly run GitHub Action for unit tests with CPU.
• Moved CLI scripts from /src/bin/**<SCRIPT>** to /src/flat_bug/cli/<SCRIPT> submodule. (and change pyproject.toml to use the new path)
• Improved README installation guide, particularly with steps related to PyTorch integration with uv/pip. (https://docs.astral.sh/uv/guides/integration/pytorch/#configuring-accelerators-with-optional-dependencies)
• Change all use of @torch.compile to @torch.compile(fullgraph=True, mode="max-autotune-no-cudagraphs").
• Add torch.set_float32_matmul_precision("medium") to the start of flat_bug.cli.fb_predict.py:main i.e. the main CLI API for this package.

Full Changelog: https://github.com/darsa-group/flat-bug/compare/v1.1-beta.0...v1.1-beta.1