Benson Ma

Results 59 issues of Benson Ma

Summary: Original commit changeset: c8c4e6fff8ef - D73693209 seems to have caused regressions on the backward adagrad test for AMD, though it passes without issue on NVIDIA. Original Phabricator Diff: D73693209...

fb-exported
cla signed

Summary: - Simplify weight row cache load and evict routines Reviewed By: sryap Differential Revision: D73693209

fb-exported
cla signed

Summary: - Fix `int32_t` to `auto` for code around `WeightRow` - Fix `kINT8QparamsBytes` from `float` to `int32_t` Reviewed By: spcyppt, sryap Differential Revision: D73690651

fb-exported
cla signed

Summary: - Migrate Jinja `make_pta_acc_format()` from the old `MAKE_PTA_WITH_NAME` and `MAKE_PTA_ACC_WITH_NAME` to using `PTA_B` and `PTA_ACC_B` Reviewed By: sryap Differential Revision: D73417820

fb-exported
cla signed

Summary: X-link: https://github.com/facebookresearch/FBGEMM/pull/1108 https://github.com/facebookresearch/FBGEMM/pull/1107 Reviewed By: q10 Differential Revision: D73622324

fb-exported
cla signed

Summary: - Is it even possible to be more thorough? #buildall - Be so, so thorough #buildmore - Be thorough #buildsonlynotests - No runtime effects! - If you approve of...

fb-exported
cla signed

- Add Nova workflow for torch 2.6 compatible releases. This is for generating nightly release for use with vLLM, which require compatibility with a lower version of PyTorch.

cla signed

- Add more docs scaffolding for GenAI

cla signed

Summary: LLVM-15 has a warning `-Wunused-variable` which we treat as an error because it's so often diagnostic of a code issue. Unused variables can compromise readability or, worse, performance. This...

fb-exported
cla signed

Summary: - Update FBGEMM versioning to 1.5.0 Differential Revision: D89254294

fb-exported
cla signed
module: rocm
meta-exported