Skip to content

Fix the GPU multi host segmentation fault on XPK #129

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

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix the GPU multi host segmentation fault on XPK #129

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a0b1e55
Update the huggingface_hub version in setup script so that pip instal…
wang2yn84 Oct 25, 2024
07563ba
Fix the GPU multi host segmentation fault on XPK.
wang2yn84 Oct 25, 2024
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
2 changes: 1 addition & 1 deletion setup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -97,7 +97,7 @@
"filelock",
"flax>=0.4.1",
"hf-doc-builder>=0.3.0",
"huggingface-hub>=0.13.2",
"huggingface-hub==0.24.7",
"requests-mock==1.10.0",
"importlib_metadata",
"invisible-watermark>=0.2.0",
Expand Down
5 changes: 1 addition & 4 deletions src/maxdiffusion/max_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -397,8 +397,7 @@ def setup_initial_state(
if not state:
max_logging.log(f"Could not find the item in orbax, creating state...")

init_train_state_partial = functools.partial(
init_train_state,
state = init_train_state(
model=model,
tx=tx,
weights_init_fn=weights_init_fn,
Expand All @@ -407,8 +406,6 @@ def setup_initial_state(
eval_only=False,
)

state = jax.jit(init_train_state_partial, in_shardings=None, out_shardings=state_mesh_shardings)()
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@wang2yn84 the jit function here shards the models according to the sharding specification in the config yml. In DDP, this won't make a difference since the state is replicated, but removing this function would prevent running FSDP or other sharding configurations.


state = unbox_logicallypartioned_trainstate(state)

return state, state_mesh_shardings
Expand Down
Loading