Repository navigation

All pull requests

FSDP: Added fix + Regression test for incorrect params_group mapping in optimizer state sharding
#1202Shasha-17 opened Dec 26, 20256 tasks
2
Fix failure in test_checkpoint_activations_norm.py
#1201maksverver opened Sep 13, 202512 tasks
Fix test failures caused by warns(None).
#1200maksverver opened Sep 13, 202512 tasks
Add support for free-threaded CPython 3.13t
#1198andfoy opened May 5, 20252 of 10 tasks
1
support for grad acc
#1190ngoyal2707 opened Aug 30, 2024Contributor
2
Bump scikit-learn from 1.1.3 to 1.5.0
#1187dependabot[bot] opened Jun 18, 2024Bot
FP8 AllGather Support in Fairscale
#1185levendlee opened May 20, 2024Member10 tasks
6
Llama4 FP8 Training Debug - fairscale
#1183jiecaoyu opened May 7, 202410 tasks
Fix minor grammatical corrections in docs
#1181aakashapoorv opened Apr 29, 20246 of 10 tasks
[FSDPv1] Only perform cat() during last microbatch backward() within FlattenParamsWrapper
#1180chrisxcai opened Apr 29, 2024
5
[WIP] Make FSDPv1 only perform cat() during last microbatch backward() within FlattenParamsWrapper
#1178chrisxcai opened Apr 19, 2024
2
Make sure that tensor is contiguous before gathering across processes
#1173patrickvonplaten opened Apr 1, 2024
Bump black from 22.3.0 to 24.3.0
#1169dependabot[bot] opened Mar 20, 2024Bot
Fairscale support for only performing allreduce in last microbatch
#1168jiecaoyu opened Mar 18, 202410 tasks
3
Fix params_with_grad in FSDP when the model has frozen parameters
#1167whbldhwj opened Mar 4, 20241 of 10 tasks
Avoid calling _free_fp16_param_shard() too early
#1164jiecaoyu opened Feb 20, 202410 tasks
3
Add support for torch.set_default_device when initializing model parameters
#1158fshp971 opened Dec 28, 20236 of 10 tasks
1
Fix _free_full_params()
#1149hadasah opened Nov 21, 202310 tasks
Extend CheckpointFunction to track all tensor input/output
#1148000Justin000 opened Nov 7, 20236 of 10 tasks
1
[Not for merge] fp8allgather debug
#1147jiecaoyu opened Oct 29, 2023
Add main grad before fwd pass
#1142vedanuj opened Oct 4, 2023
10
Add main_grad
#1140jianyuh opened Oct 2, 2023Member10 tasks
6
Fp8 all gather hack
#1136jspark1105 opened Sep 17, 2023
6
Fix typo in ParallelEmbedding argument assignment
#1132hessamb opened Aug 28, 2023
4
Add a context manager for activation sharding.
#1127luyug opened Jun 12, 2023
1

Navigation Menu

Repository navigation

All pull requests

FSDP: Added fix + Regression test for incorrect params_group mapping in optimizer state sharding

Fix failure in test_checkpoint_activations_norm.py

Fix test failures caused by `warns(None)`.

Add support for free-threaded CPython 3.13t

support for grad acc

Bump scikit-learn from 1.1.3 to 1.5.0

FP8 AllGather Support in Fairscale

Llama4 FP8 Training Debug - fairscale

Fix minor grammatical corrections in docs

[FSDPv1] Only perform cat() during last microbatch backward() within FlattenParamsWrapper

[WIP] Make FSDPv1 only perform cat() during last microbatch backward() within FlattenParamsWrapper

Make sure that tensor is contiguous before gathering across processes

Bump black from 22.3.0 to 24.3.0

Fairscale support for only performing allreduce in last microbatch

Fix params_with_grad in FSDP when the model has frozen parameters

Avoid calling _free_fp16_param_shard() too early

Add support for `torch.set_default_device` when initializing model parameters

Fix _free_full_params()

Extend CheckpointFunction to track all tensor input/output

[Not for merge] fp8allgather debug

Add main grad before fwd pass

Add main_grad

Fp8 all gather hack

Fix typo in ParallelEmbedding argument assignment

Add a context manager for activation sharding.

Pull requests