Skip to content

Migrate fsdp1 -> fsdp2 in Tensor Parallelism Tutorial #3464

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jul 16, 2025
Merged

Migrate fsdp1 -> fsdp2 in Tensor Parallelism Tutorial #3464

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions intermediate_source/TP_tutorial.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -333,7 +333,7 @@ This 2-D parallelism pattern can be easily expressed via a 2-D DeviceMesh, and w

from torch.distributed.device_mesh import init_device_mesh
from torch.distributed.tensor.parallel import ColwiseParallel, RowwiseParallel, parallelize_module
from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
from torch.distributed.fsdp import fully_shard

# i.e. 2-D mesh is [dp, tp], training on 64 GPUs that performs 8 way DP and 8 way TP
mesh_2d = init_device_mesh("cuda", (8, 8))
Expand All @@ -347,7 +347,7 @@ This 2-D parallelism pattern can be easily expressed via a 2-D DeviceMesh, and w
# apply Tensor Parallel intra-host on tp_mesh
model_tp = parallelize_module(model, tp_mesh, tp_plan)
# apply FSDP inter-host on dp_mesh
model_2d = FSDP(model_tp, device_mesh=dp_mesh, use_orig_params=True, ...)
model_2d = fully_shard(model_tp, mesh=dp_mesh, ...)


This would allow us to easily apply Tensor Parallel within each host (intra-host) and apply FSDP across hosts (inter-hosts), with **0-code changes** to the Llama model.
Expand Down