8000 Commits · tombousso/pytorch · GitHub

More Web Proxy on the site http://driver.im/

Commits

Commits on Jun 8, 2024

[fsdp2] simplify fsdp_param logic with DTensorSpec (#128242)

wanchaol
authored and
pytorchmergebot
committed
[fsdp2] update foreach_reduce accumulate_grad (#128117)

wanchaol
authored and
pytorchmergebot
committed

Commits on Jun 7, 2024

[dtensor] directly return local_tensor under no_grad (#128145)

wanchaol
authored and
pytorchmergebot
committed

Commits on Jun 6, 2024

[dtensor] reuse DTensorSpec as much as possible (#128112)

wanchaol
authored and
pytorchmergebot
committed

Commits on Jun 3, 2024

[dtensor] implement scatter op with simple replication (#126713)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] standardize multi mesh-dim strategy with utils (#126712)

wanchaol
authored and
pytorchmergebot
committed

Commits on May 29, 2024

Commits on May 22, 2024

[dtensor] implement distributed topk operator (#126711)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] remove unused failed_reason (#126710)

wanchaol
authored and
pytorchmergebot
committed
[SAC] fix ignored ops in eager mode to recompute (#126751)

wanchaol
authored and
pytorchmergebot
committed

Commits on May 16, 2024

fix the device type for with_comms decorator (#125798)

wanchaol
authored and
pytorchmergebot
committed

Commits on May 14, 2024

make sure device mesh can be imported from torch.distributed (#126119)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] early return for _split_tensor (#125810)

wanchaol
authored and
pytorchmergebot
committed

Commits on May 9, 2024

[fsdp2] Accomodate FSDP2 to accept parent mesh > 2 (#125778)

wanchaol
authored and
pytorchmergebot
committed

Commits on May 8, 2024

[tp] add some test for shard output layouts for rowwise parallel (#125713)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] improve some pretty print in op schema (#125695)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] make sure meta tensor random op does not alternate rng state (#125693)

wanchaol
authored and
pytorchmergebot
committed

Commits on May 6, 2024

try to fix the warning in distribute_tensor (#125476)

wanchaol
authored and
pytorchmergebot
committed

Commits on May 4, 2024

try to fix the warning in distribute_tensor (#125476)

wanchaol
authored and
pytorchmergebot
committed
[comm_mode] adding some initial c10d ops to CommDebugMode (#125475)

wanchaol
authored and
pytorchmergebot
committed

Commits on Apr 30, 2024

[dtensor] delete the old unused mesh_alltoall (#124879)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] implement shard dim change with alltoall (#124872)

wanchaol
authored and
pytorchmergebot
committed

Commits on Apr 29, 2024

Commits on Apr 25, 2024

[dtensor] move pad/unpad_tensor to separate utils (#124871)

wanchaol
authored and
pytorchmergebot
committed

Commits on Apr 24, 2024

[device_mesh] add a private init backend option (#124780)

wanchaol
authored and
pytorchmergebot
committed

Commits on Apr 22, 2024

[tp] add kwargs support to prepare_module_input (#124114)

wanchaol
authored and
pytorchmergebot
committed

Commits on Apr 1, 2024

[dtensor] switch aten.t to use op strategy (#122950)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] make sure expected input spec have correct tensor meta (#122949)

wanchaol
authored and
pytorchmergebot
committed
[dtensor] refactor schema suggestions in output sharding (#122929)

wanchaol
authored and
pytorchmergebot
committed

0