-
Notifications
You must be signed in to change notification settings - Fork 476
Pull requests: pytorch/torchtitan
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
[DSV3] Remove keep a copy of GroupedExperts weight, free memory in StateDictAdapter
CLA Signed
This label is managed by the Meta Open Source bot.
#1585
opened Aug 16, 2025 by
wwwjn
Loading…
Muon with 3D tensors
CLA Signed
This label is managed by the Meta Open Source bot.
#1584
opened Aug 16, 2025 by
byronxu99
Loading…
Fix typos
CLA Signed
This label is managed by the Meta Open Source bot.
#1583
opened Aug 16, 2025 by
BioGeek
Loading…
[HF] Model Definition Conversion Support for FLUX
CLA Signed
This label is managed by the Meta Open Source bot.
#1582
opened Aug 15, 2025 by
wesleytruong
Loading…
[demo][fsdp2][ep] explicit prefetching to overlap all-gather with cuda sync
CLA Signed
This label is managed by the Meta Open Source bot.
#1581
opened Aug 15, 2025 by
weifengpy
Loading…
Add config to AC to toggle early-stop
CLA Signed
This label is managed by the Meta Open Source bot.
#1580
opened Aug 15, 2025 by
soulitzer
Loading…
feat: get_extra_metrics
CLA Signed
This label is managed by the Meta Open Source bot.
#1578
opened Aug 15, 2025 by
garrett361
Loading…
[EP] add initial support for NVSHMEM-based all-to-all
CLA Signed
This label is managed by the Meta Open Source bot.
#1569
opened Aug 14, 2025 by
tianyu-l
Loading…
[Do Not Land] Debug for SDPA + CP nan issue in DeepSeekV3
CLA Signed
This label is managed by the Meta Open Source bot.
Multinode SkyPilot example
CLA Signed
This label is managed by the Meta Open Source bot.
#1564
opened Aug 13, 2025 by
alex000kim
Loading…
fix: remove redundant legacy usage of mp in checkpoint
CLA Signed
This label is managed by the Meta Open Source bot.
#1562
opened Aug 13, 2025 by
yzs981130
Loading…
[WIP] Experimental implementation of gpt-oss (grouped GEMM MoE + FlexAttention sink/sliding)
#1559
opened Aug 13, 2025 by
KhoomeiK
Loading…
[PoC] Enable flexible different layout for same mesh via a util function
CLA Signed
This label is managed by the Meta Open Source bot.
#1550
opened Aug 11, 2025 by
fduwjj
Loading…
[WIP] [mxfp8] torchao mxfp8 moe integration
CLA Signed
This label is managed by the Meta Open Source bot.
#1549
opened Aug 11, 2025 by
danielvegamyhre
•
Draft
added example for bidirectional checkpoint testing
CLA Signed
This label is managed by the Meta Open Source bot.
#1540
opened Aug 6, 2025 by
wesleytruong
Loading…
add support for simplefsdp+ep
CLA Signed
This label is managed by the Meta Open Source bot.
#1529
opened Aug 5, 2025 by
ruisizhang123
Loading…
Adding logic for cleaning up FT checkpoints
CLA Signed
This label is managed by the Meta Open Source bot.
#1528
opened Aug 5, 2025 by
bentherien
Loading…
[WIP][Dion Official Optimizer, Muon] Integrate official Dion, and high speed Muon, optimizer impl with TorchTitan and Optimizer component class
CLA Signed
This label is managed by the Meta Open Source bot.
Fix semi-sync training with 1GPU per FT replica
CLA Signed
This label is managed by the Meta Open Source bot.
#1505
opened Jul 31, 2025 by
bentherien
Loading…
perf testing
CLA Signed
This label is managed by the Meta Open Source bot.
#1488
opened Jul 29, 2025 by
ankitageorge
•
Draft
[Evaluation] Adding evaluation feature to TorchTitan
CLA Signed
This label is managed by the Meta Open Source bot.
#1470
opened Jul 28, 2025 by
raymin0223
Loading…
[autoparallel] Enable bucketing passes for autoparallel, reorder and sink_waits.
CLA Signed
This label is managed by the Meta Open Source bot.
#1463
opened Jul 25, 2025 by
IvanKobzarev
Loading…
Autoparallel support for DP-only, DP+TP, or TP-only
CLA Signed
This label is managed by the Meta Open Source bot.
#1459
opened Jul 25, 2025 by
IvanKobzarev
Loading…
[WIP] Integrate autoparallel into torchtitan
CLA Signed
This label is managed by the Meta Open Source bot.
#1458
opened Jul 25, 2025 by
IvanKobzarev
Loading…
add lr logging
CLA Signed
This label is managed by the Meta Open Source bot.
#1453
opened Jul 24, 2025 by
samsja
Loading…
Previous Next
ProTip!
Filter pull requests by the default branch with base:main.