13 Pull requests merged by 3 people

• Refactor local_sgd integration tests

  #96 merged Feb 4, 2025

• Participants APIs should check if quorum is started

  #95 merged Feb 4, 2025

• manager: expose participating_rank

  #94 merged Feb 3, 2025

• examples,docs: adjust ddp example timeout and docs

  #93 merged Jan 31, 2025

• Add DiLoCo

  #92 merged Jan 31, 2025

• ProcessGroupBabyNCCL: support multiple streams and use event on start

  #91 merged Jan 31, 2025

• Add DiLoCo

  #76 merged Jan 30, 2025

• CheckpointServer: start in disallowed state + tests

  #90 merged Jan 30, 2025

• Improve OptimizerWrapper composability

  #85 merged Jan 30, 2025

• Change how TorchFT manages user_state_dict

  #87 merged Jan 30, 2025

• ProcessGroupBaby: support full suite of PG tests

  #89 merged Jan 29, 2025

• Fix ManagedDeviceMesh composability issues

  #86 merged Jan 29, 2025

• process_group: fix docs with torch==2.6.0

  #88 merged Jan 29, 2025

2 Issues closed by 1 person

• ManagerClient.quorum should return a namedtuple, dataclass or object

  #52 closed Feb 4, 2025

• [lighthouse] use heartbeat info to quickly drop down replicas

  #35 closed Feb 4, 2025

1 Issue opened by 1 person

• process_group: support all PG APIs

  #97 opened Feb 4, 2025

3 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.

• LocalSGD / DiLoCo support

  #39 commented on Feb 4, 2025 • 0 new comments

• [WIP] FSDP example

  #77 commented on Feb 4, 2025 • 0 new comments

• [WIP][RFC] Required changes for integration with TorchTitan

  #82 commented on Jan 29, 2025 • 0 new comments