Pull requests: pytorch/torchtitan
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Fixes for manual stage shape and pp_degree=3, WIP
CLA Signed
This label is managed by the Meta Open Source bot.
#340
opened May 17, 2024 by
wconstab
Loading…
only produce tensorboard logs on rank 0 by default
CLA Signed
This label is managed by the Meta Open Source bot.
#339
opened May 16, 2024 by
tianyu-l
Loading…
Test 1f1b schedule
CLA Signed
This label is managed by the Meta Open Source bot.
#337
opened May 16, 2024 by
wconstab
Loading…
Add 8gpu runner
CLA Signed
This label is managed by the Meta Open Source bot.
#327
opened May 15, 2024 by
wconstab
Loading…
selective compilation - norm layers only
CLA Signed
This label is managed by the Meta Open Source bot.
#320
opened May 10, 2024 by
lessw2020
Loading…
Add support of DDP and CompiledAutograd.
CLA Signed
This label is managed by the Meta Open Source bot.
#319
opened May 9, 2024 by
fegin
Loading…
Add Pipeline Parallel (and 2D PP+FSDP) support
CLA Signed
This label is managed by the Meta Open Source bot.
#318
opened May 9, 2024 by
wconstab
Loading…
[fused_rmsnorm] Register as a custom operator for tracing
CLA Signed
This label is managed by the Meta Open Source bot.
#303
opened May 3, 2024 by
wconstab
Loading…
[fused_rmsnorm] Avoid querying device inside forward
CLA Signed
This label is managed by the Meta Open Source bot.
#301
opened May 3, 2024 by
wconstab
Loading…
[fused_rmsnorm] Avoid conditional on dynamic stride
CLA Signed
This label is managed by the Meta Open Source bot.
#300
opened May 3, 2024 by
wconstab
Loading…
register fused rmsnorm as pytorch custom op
CLA Signed
This label is managed by the Meta Open Source bot.
[wip] differentiate Rstd vs rstd
CLA Signed
This label is managed by the Meta Open Source bot.
#294
opened May 2, 2024 by
lessw2020
Loading…
Use stateful dataloader to checkpoint data iteration order and token buffer
CLA Signed
This label is managed by the Meta Open Source bot.
#279
opened Apr 26, 2024 by
gokulavasan
Loading…
torch.compile each TransformerBlock instead of the whole model
CLA Signed
This label is managed by the Meta Open Source bot.
#268
opened Apr 25, 2024 by
wanchaol
Loading…
RFC for ckpt apis
CLA Signed
This label is managed by the Meta Open Source bot.
#226
opened Apr 13, 2024 by
wconstab
Loading…
[RFC] Sharded embeddings in separate FSDP group
CLA Signed
This label is managed by the Meta Open Source bot.
run sdpa with dtensor
CLA Signed
This label is managed by the Meta Open Source bot.
#180
opened Mar 30, 2024 by
tianyu-l
Loading…
Implement fast checkpoint path
CLA Signed
This label is managed by the Meta Open Source bot.
#127
opened Mar 12, 2024 by
fegin
Loading…
ProTip!
Exclude everything labeled
bug
with -label:bug.