12 Pull requests opened by 10 people

• Fix torchrec type error

  #2973 opened May 16, 2025

• Revert D74293458

  #2974 opened May 16, 2025

• split train_pipeline.utils - tracing

  #2982 opened May 19, 2025

• mark metricmodule methods as experimental

  #2983 opened May 19, 2025

• Refactoring ITEP / PTP Pruning Scuba Logger [1/N]

  #2986 opened May 21, 2025

• Support RecMetrics new options in Pyper config

  #2987 opened May 21, 2025

• Fix undeclared type in colab for sharding examples

  #2989 opened May 21, 2025

• Enable fx tracing for maybe_td_to_kjt

  #2991 opened May 21, 2025

• Harden Tower QPS Fused Mode

  #2992 opened May 21, 2025

• Update cuda versions for CI tests

  #2994 opened May 22, 2025

• split train_pipeline.utils - forward

  #2995 opened May 22, 2025

• Migrate to correct logger interface

  #2996 opened May 23, 2025

14 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.

• `AttributeError: 'NoneType' object has no attribute '_dynamo_weak_dynamic_indices'` when using row-wise sharding

  #1890 commented on May 18, 2025 • 0 new comments

• Compatibility with PyTorch 2.7.0 and FBGEMM_GPU 1.2.0?

  #2971 commented on May 22, 2025 • 0 new comments

• Modify torchrec code to use reduce_scatter_v and all_gather_v

  #568 commented on May 17, 2025 • 0 new comments

• fix state_dict compatibility with ebc family

  #703 commented on May 17, 2025 • 0 new comments

• create fx argument for sparsefeature types

  #708 commented on May 17, 2025 • 0 new comments

• fix dmp copy

  #1100 commented on May 17, 2025 • 0 new comments

• wrapping to jt_dict

  #1306 commented on May 17, 2025 • 0 new comments

• feature-batched inference QEC

  #1491 commented on May 17, 2025 • 0 new comments

• static config for sharded_modules

  #1766 commented on May 17, 2025 • 0 new comments

• add bag reverse mapping for block_bucketize kernel

  #1845 commented on May 17, 2025 • 0 new comments

• Back out "Replace LR access with wrapper"

  #2839 commented on May 17, 2025 • 0 new comments

• Allow a context manager to be called around apply_jit

  #2927 commented on May 16, 2025 • 0 new comments

• DistributedModelParallel resharding Interface

  #2945 commented on May 21, 2025 • 0 new comments

• convert stride_per_key_per_rank to tensor inside KJT

  #2959 commented on May 21, 2025 • 0 new comments