| name | torchfits-dev |
| description | Develop torchfits (FITS I/O for PyTorch) — tensor/table contracts, C++ CFITSIO engine, verify tiers, docs contract |
| metadata | {"project":"torchfits","stack":"python, c++, pytorch, pixi"} |
torchfits development
FITS I/O for PyTorch: read/write FITS as tensors and tables, cutouts, predicate
filters, and a shell CLI, backed by a C++ engine with vendored CFITSIO. GPU
acceleration (CUDA on Linux, MPS on macOS) when available. See AGENTS.md.
Verify tiers
| When | Command |
|---|
| During edits | pixi run preflight-push |
| Before push / PR | pixi run ci-local |
| Before tag | pixi run release-gate |
Contract rules
- Docs must match the public façade (
docs/api*.md); env vars must exist in
src/. Run pixi run docs-contract and pixi run docs-links on docs changes.
- User-facing docs: human-first, copy-pasteable recipes; zero internal jargon.
- Correctness is paramount: parity against the CFITSIO/public API contract.
- Prefer smallest correct diffs; no new dependencies without a clear need.
- Performance claims need before/after timing from an existing
pixi run bench-*
case (same host, same case_id).
Surface
torchfits.read_tensor("img.fits", device="cuda") / write / open (MEF)torchfits.table.read(..., where="MAG < 20") / read_torchFitsImageDataset + make_loader(...)- CLI:
torchfits info / header / convert / cutout / …
