-
Notifications
You must be signed in to change notification settings - Fork 791
Pull requests: pytorch/torchtitan
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
fix(hf_datasets): shuffle HuggingFaceTextDataset on re-loop and replay on resume
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3023
opened Apr 18, 2026 by
CrepuscularIRIS
Loading…
[Do Not Merge] DUMMY PR: Find Device Name for MI325 runner
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3021
opened Apr 18, 2026 by
akashveramd
Collaborator
•
Draft
[GraphTrainer][AutoDev] Remove compile_with_inductor annotation from qwen3 FlexAttention
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3019
opened Apr 17, 2026 by
SherlockNoMad
Contributor
•
Draft
3 of 4 tasks
[GraphTrainer][AutoDev] Remove unused standalone GraphTrainerConfig dataclass
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3018
opened Apr 17, 2026 by
SherlockNoMad
Contributor
•
Draft
3 of 4 tasks
Gate H100.8 CI workflows behind ciflow/h100.8 label
ciflow/h100.8
Trigger H100.8 CI
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3016
opened Apr 17, 2026 by
SherlockNoMad
Contributor
Loading…
3 tasks
[GraphTrainer] Introduce CPU offload pass for activation memory savings
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3015
opened Apr 17, 2026 by
mlazos
Loading…
[graph_trainer] Add MANIFESTO.md
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3014
opened Apr 17, 2026 by
SherlockNoMad
Contributor
Loading…
fix: normalize n_tokens_seen by cp_degree when context parallelism is…
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3009
opened Apr 17, 2026 by
TryingtobeingNikhil
Loading…
Fix: reproducible training resume across epoch boundaries for map and streaming datasets
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3008
opened Apr 17, 2026 by
slimfrkha
Contributor
Loading…
[HybridEP] Enable HybridEP with graph_trainer AOT compilation
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3007
opened Apr 17, 2026 by
syed-ahmed
Contributor
Loading…
1 task
[ignore-for-now][llm_trainer] Add experiment for LLM-driven model optimization
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3006
opened Apr 17, 2026 by
bobrenjc93
Contributor
•
Draft
[rl] [not ready for review] Generator refactor
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#3001
opened Apr 16, 2026 by
joecummings
Member
•
Draft
Fix DTensor attr handling in make_fx tracer
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2998
opened Apr 16, 2026 by
tugsbayasgalan
Contributor
Loading…
[graph_trainer] Copy forward metadata to backward subgraphs
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2995
opened Apr 16, 2026 by
tugsbayasgalan
Contributor
Loading…
Fix batch invariant mode: using NCCL tree based all-reduce
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2994
opened Apr 16, 2026 by
wwwjn
Contributor
Loading…
[NOT READY FOR REVIEW][Full DTensor] Config-based Full DTensor for Llama3
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
[NOT READY FOR REVIEW][Module] Remove LocalMapInnerAttention, use static LocalMapSpec
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2986
opened Apr 15, 2026 by
fegin
Contributor
Loading…
[rl] Trainer refactor
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2985
opened Apr 15, 2026 by
joecummings
Member
Loading…
Refactor Loss with Scale to fix PP last stage gradients
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2984
opened Apr 15, 2026 by
wwwjn
Contributor
Loading…
Increase timeout for features test to 60 minutes.
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2982
opened Apr 15, 2026 by
akashveramd
Collaborator
Loading…
re-enable compile tests
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2978
opened Apr 15, 2026 by
acisseJZhong
Contributor
Loading…
[demo] verify fully_shard([norm, head]) and fully_shard([tok_embedding, norm, head]) works with chunked loss
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
Collect peak memory directly in integration tests
ciflow/8gpu
CLA Signed
This label is managed by the Meta Open Source bot.
#2974
opened Apr 15, 2026 by
tugsbayasgalan
Contributor
Loading…
Previous Next
ProTip!
Filter pull requests by the default branch with base:main.