-
Notifications
You must be signed in to change notification settings - Fork 699
Pull requests: pytorch/torchtitan
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[DRAFT] Consolidate simple_fsdp and compiler_toolkit experiments
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2360
opened Feb 10, 2026 by
yiming0416
•
Draft
[Bugfix] Fix This label is managed by the Meta Open Source bot.
simple_rl_multiprocess.py to be runnable with recent vLLM version
ciflow/8gpu
CLA Signed
#2359
opened Feb 10, 2026 by
Lucaskabela
Loading…
[Bugfix] Fix bitwise determinism after vLLM SiluAndMul change
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2358
opened Feb 9, 2026 by
Lucaskabela
Loading…
[SAC] Refactor activation checkpointing to use centralized policy-based approach
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
[RFC][DONT LAND] Support different state_dict for save and load
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
[ci] Add DSv3 SimpleFSDP auto_bucketing to h100 ci jobs
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2347
opened Feb 9, 2026 by
IvanKobzarev
Loading…
[simple_fsdp] Use schedule_overlap_bucketing_from_inductor_configs for overlap_passes
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2346
opened Feb 9, 2026 by
IvanKobzarev
Loading…
[DRAFT] Optimize MoE Routing via This label is managed by the Meta Open Source bot.
torch.sort Indices DType Injection
CLA Signed
[dsv3] per-layer error when compile with MoE "HOP: Unsafe side effect"
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2341
opened Feb 7, 2026 by
weifengpy
Loading…
Add run-to-run determinism testing to H100 CI
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2339
opened Feb 6, 2026 by
xmfan
Loading…
random_experiment
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2335
opened Feb 6, 2026 by
anshul-si
Loading…
[torchcomms] Simplify ParallelDims to use base class inheritance and mesh views
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2334
opened Feb 6, 2026 by
mori360
Loading…
Torchtitan changes to integrate into Verl
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2333
opened Feb 5, 2026 by
acisseJZhong
Loading…
Implement sharding and device mesh debug tool
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2328
opened Feb 5, 2026 by
fegin
Loading…
Disable DDP averaging to avoid repeated gradient averaging
bug
Something isn't working
CLA Signed
This label is managed by the Meta Open Source bot.
fb-exported
meta-exported
#2323
opened Feb 4, 2026 by
Shagun-G
Loading…
Apply #1895 only when really necessary
CLA Signed
This label is managed by the Meta Open Source bot.
#2322
opened Feb 4, 2026 by
ericschreiber
Loading…
Fixed autoparallel integration tests on ROCm.
CLA Signed
This label is managed by the Meta Open Source bot.
module: rocm
#2321
opened Feb 4, 2026 by
wenchenvincent
Loading…
[No Merge] Debug autoparallel test ci
CLA Signed
This label is managed by the Meta Open Source bot.
#2317
opened Feb 3, 2026 by
wenchenvincent
•
Draft
Register This label is managed by the Meta Open Source bot.
_ScaledPartial placement
CLA Signed
#2313
opened Feb 2, 2026 by
Aidyn-A
Loading…
separate out training for fault tolerance
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2311
opened Feb 2, 2026 by
tushar00jain
Loading…
Add Transformer-Engine Fused_Adam Optimizer Support
CLA Signed
This label is managed by the Meta Open Source bot.
[draft][lora] Apply LoraLinear as a wrapper of Linear
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
[FSDP2] enable per-param mesh FSDP2 for MoE
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
[DeepEP Integration] Free cache after combine for forward only path.
CLA Signed
This label is managed by the Meta Open Source bot.
Enable graph_pp for autoparallel in torchtitan
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2271
opened Jan 23, 2026 by
sanketpurandare
•
Draft
Previous Next
ProTip!
no:milestone will show everything without a milestone.