Skip to content

Pull requests: pytorch/FBGEMM

New pull request New
598 Open 4,441 Closed
598 Open 4,441 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Cleanup branches for CUDA 9 cla signed
#5269 opened Dec 21, 2025 by cyyever Loading…
Add tidy fixes cla signed
#5268 opened Dec 21, 2025 by cyyever Loading…
Replace .data_ptr with .mutable_data_ptr or .const_data_ptr cla signed
#5267 opened Dec 20, 2025 by cyyever Loading…
Optimizations for index_select_scalar_cumsum_kernel on ROCm cla signed module: rocm
#5263 opened Dec 18, 2025 by amd-wsung102 Loading…
3
Refactor TBE benchmark reporter to use structured data config cla signed fb-exported meta-exported
#5260 opened Dec 18, 2025 by gchalump Loading…
1
Fix blackwell CUTLASS attention meta registration + actually test compile cla signed fb-exported meta-exported
#5259 opened Dec 18, 2025 by jbschlosser Loading…
1
Optimize benchmark index generation with std::sample() cla signed fb-exported meta-exported
#5254 opened Dec 17, 2025 by terdogan Loading…
1
Remove unused dedup_map and associated includes from benchmarks cla signed fb-exported meta-exported
#5253 opened Dec 17, 2025 by terdogan Loading…
1
Move the prefetched info to preallocated buffers cla signed fb-exported meta-exported
#5251 opened Dec 17, 2025 by chouxi Loading…
1
Enable direct MX4→BF16 dequantization to reduce memory (python side) (2/2) cla signed fb-exported meta-exported
#5250 opened Dec 17, 2025 by armandsauzay Loading…
1
Add aarch64 intrinsic-based dequantization to autovec routine cla signed fb-exported meta-exported
#5249 opened Dec 17, 2025 by Nicoshev Loading…
1
Choose _autovec version of GenerateEmbeddingSpMDMRowWiseSparse on AArch64 cla signed fb-exported meta-exported
#5247 opened Dec 17, 2025 by MatzeB Loading…
1
Specialize more cases to improve EmbeddingSpMDMNBitBenchmark cla signed fb-exported meta-exported
#5245 opened Dec 17, 2025 by MatzeB Loading…
1
Add EmbeddingSpMDMNBitRowWiseSparse autovectorized variant cla signed fb-exported meta-exported
#5244 opened Dec 17, 2025 by MatzeB Loading…
1
Optimize group_index_select_or_add_2d_kernel on ROCm by adding a separate codepath for small embedding dimensions cla signed module: rocm
#5233 opened Dec 16, 2025 by aryaman-gupta Loading…
1
support object cache in ssd l2 cache and add more unit tests cla signed fb-exported meta-exported
#5228 opened Dec 16, 2025 by zhaojuanmao Loading…
1
Optimizing 4-bit dequant to FP32 on AArch64 using vectorized intrinsics in EmbeddingSpMDMAutovec cla signed
#5224 opened Dec 15, 2025 by marma01 Loading…
Update heuristic to support variant batch sizes cla signed fb-exported meta-exported
#5211 opened Dec 10, 2025 by zjing14 Loading…
1
Use H100 runners for OSS CI cla signed fb-exported meta-exported
#5205 opened Dec 9, 2025 by q10 Loading…
1
Modifying clear_all_staged_data to accomadate KV Tensor Deletion cla signed fb-exported meta-exported
#5202 opened Dec 9, 2025 by Raahul46 Loading…
1
creating delete_rocksdb_checkpoint_dir function under KV Tensor cla signed fb-exported meta-exported
#5201 opened Dec 9, 2025 by Raahul46 Loading…
1
Adding returnKVTensorMetaData flag to Staging Read Strategy cla signed fb-exported meta-exported
#5200 opened Dec 9, 2025 by Raahul46 Loading…
1
Remove unused AVX{2,512}_FLAGS cla signed
#5198 opened Dec 9, 2025 by cyyever Loading…
3
Fix jagged_to_padded_dense autograd cla signed fb-exported meta-exported
#5191 opened Dec 8, 2025 by yunjiangster Loading…
1
Add warp parallelism to populate_bucketized_permute cla signed fb-exported meta-exported
#5189 opened Dec 8, 2025 by AlbertDachiChen Loading…
1
ProTip! Follow long discussions with comments:>50.