35 Pull requests opened by 20 people

• Use an env variable to roll out model sharding cache

  #2764 opened Feb 25, 2025

• Support model buffers as pipeline postproc inputs (#2744)

  #2769 opened Feb 27, 2025

• add WSD learning rate scheduler

  #2771 opened Feb 27, 2025

• Reset padding to default if not matching the qcoom type

  #2772 opened Feb 27, 2025

• Simplify batched_embedding_kernel (#2774)

  #2779 opened Mar 5, 2025

• rw sharding

  #2780 opened Mar 6, 2025

• Back out "fix flaky test due to input_jkt.weight dtype"

  #2783 opened Mar 6, 2025

• reland D70126859

  #2787 opened Mar 7, 2025

• fix validate nightly binaries for CUDA 12.6 and 12.8

  #2792 opened Mar 10, 2025

• OSS ZCH Kernels

  #2793 opened Mar 10, 2025

• support quantize fp32 embedding tables

  #2794 opened Mar 11, 2025

• change | to Union for py 3.9 compat

  #2795 opened Mar 11, 2025

• Reduce unnecesary function call and use zip to transpose 2d list

  #2796 opened Mar 11, 2025

• DO NOT LAND

  #2798 opened Mar 11, 2025

• Fix importlib metadata

  #2800 opened Mar 11, 2025

• Temporary Commit at 3/11/2025, 3:38:30 PM

  #2801 opened Mar 11, 2025

• Temporary Commit at 3/11/2025, 3:41:43 PM

  #2802 opened Mar 11, 2025

• Temporary Commit at 3/11/2025, 3:43:27 PM

  #2803 opened Mar 11, 2025

• Support placing CW shards for same table on different devices for SQEC

  #2804 opened Mar 11, 2025

• Temporary Commit at 3/11/2025, 4:05:37 PM

  #2805 opened Mar 11, 2025

• fix row-wise alltoall error when some embeddings use mean pooling and others use sum pooling

  #2809 opened Mar 12, 2025

• Add support for compile optimizer step

  #2810 opened Mar 12, 2025

• ci: Update linux.20_04 --> linux.24_04

  #2817 opened Mar 13, 2025

• Fix symbolic tracing error during IEN lowering

  #2818 opened Mar 13, 2025

• fix static_dict_gather precision issue when use mc-ebc

  #2826 opened Mar 17, 2025

• fix divide NoneType error when use mc-ebc and mean pooling

  #2829 opened Mar 17, 2025

• [oss][ci] fix tests and cmake C++ linking

  #2834 opened Mar 19, 2025

• Add SimpleFSDP to leaf node

  #2835 opened Mar 19, 2025

• Multi forward MCH eviction fix

  #2836 opened Mar 19, 2025

• Back out "Replace LR access with wrapper"

  #2839 opened Mar 20, 2025

• Force import with post script OSS job

  #2840 opened Mar 20, 2025

• Always remove the num_batch key from the checkpoint dict

  #2841 opened Mar 21, 2025

• Fix importlib metadata for binaries

  #2843 opened Mar 21, 2025

• refactoring ModelInput generation utils

  #2844 opened Mar 22, 2025

• Enable Cuda 12.6 since FBGEMM offers it now

  #2845 opened Mar 24, 2025

3 Issues closed by 3 people

• gpu version `int_nbit_split_embedding_codegen_lookup_function` not support `indices` and `offsets` with uint64 dtype

  #1830 closed Mar 12, 2025

• [inductor] Precompilation start time is the time when a config is added to the queue, not when executor starts compiling the config

  #2775 closed Mar 7, 2025

• kaggle dataset in dlrm torchrec train occurs the always 0.5 AUROC

  #2776 closed Mar 6, 2025

8 Issues opened by 5 people

• [Bug]: ShardedManagedCollisionEmbeddingCollection throws an IndexError when "return_remapped_features=True"

  #2838 opened Mar 20, 2025

• TypeError: unsupported operand type(s) for /: 'Tensor' and 'NoneType' when use mc-ebc and mean pooling

  #2828 opened Mar 17, 2025

• The `static_dict_gather` function encounters precision issue when use mc-ebc

  #2827 opened Mar 17, 2025

• roadmap 2025 H1 H2

  #2825 opened Mar 17, 2025

• [Question] Is there any way to apply AdamW as optimizer for sparse features

  #2821 opened Mar 14, 2025

• row-wise alltoall error when some embeddings use mean pooling and others use sum pooling

  #2808 opened Mar 12, 2025

• Floating point exception (core dumped) when use quantize embeddings with float32 dtype

  #2807 opened Mar 12, 2025

• Support empty batch for EmbeddingCollection

  #2773 opened Feb 28, 2025

4 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• wrong local_world_size fallbacks to global world size with slurm

  #2667 commented on Mar 4, 2025 • 0 new comments

• question about parallelism for embedding

  #2119 commented on Mar 11, 2025 • 0 new comments

• tag KJT related part to sc.INPUT_DIST

  #2678 commented on Mar 10, 2025 • 0 new comments

• add QuantManagedCollisionEmbeddingBagCollection

  #2723 commented on Feb 28, 2025 • 0 new comments