Junjie Qi 3a98062426 faiss/conda: extend LSQ NEON-flaky test skip to all arm64 (#5713) ...

Summary:
Pull Request resolved: https://github.com/facebookresearch/faiss/pull/5713

Two Local Search Quantizer tests are numerically flaky under ARM NEON FMA ordering:
- `TestComponents_ARM_NEON.test_update_codebooks_with_double`: `err_double` can exceed `err_float` on ARM even though it is better on x86.
- `TestProductLocalSearchQuantizer.test_lut`: `compute_LUT` rounds differently on NEON; max relative diff reaches ~1.7e-3, above the x86-tuned `rtol=5e-4`.

D121930308 already skips both via `pytest -k` on `osx and arm64`. This widens the conda selectors to `arm64` so linux-aarch64 nightly builds skip them too:
- `pytest` test requirement and the `pytest -k` command: `[osx and arm64]` -> `[arm64]`
- `unittest` `test_*` command: `[not (osx and arm64)]` -> `[not arm64]`

The root fix (platform guards / looser `rtol` in the test files) is still pending.

Reviewed By: mnorris11

Differential Revision: D122238335

fbshipit-source-id: c18fa2ecec29602e758bc20e2bc0dfc3e5d207d0

2026-10-08 11:56:58 -07:00

..

faiss

faiss/conda: extend LSQ NEON-flaky test skip to all arm64 (#5713)

2026-10-08 11:56:58 -07:00

faiss-gpu

Fix SIMD level detection in Faiss (#5507)

2026-08-10 12:49:24 -07:00

faiss-gpu-cuvs

Fix SIMD level detection in Faiss (#5507)

2026-08-10 12:49:24 -07:00

conda_build_config.yaml

Update python to include 3.13 and 3.14 (#4859)

2026-03-03 17:19:33 -08:00