Skip to content

[model loading] don't gc.collect() if only 1 shard is used #36721

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
gante merged 3 commits into from
Mar 14, 2025
Merged

[model loading] don't gc.collect() if only 1 shard is used #36721

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f883448
don't gc collect if 1 shard is used
gante Mar 14, 2025
9dac0a4
delete state dict anyways
gante Mar 14, 2025
5ca357f
Merge branch 'main' into dont_gc_collect_on_single_shard
gante Mar 14, 2025
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
8 changes: 5 additions & 3 deletions src/transformers/modeling_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4831,6 +4831,7 @@ def _load_pretrained_model(

error_msgs = []
mismatched_keys = []
has_multiple_shards = len(checkpoint_files) > 1
# Iterate on all the shards to load the weights
for shard_file in checkpoint_files:
# Skip the load for shards that only contain disk-offloaded weights
Expand All @@ -4849,7 +4850,7 @@ def _load_pretrained_model(
):
map_location = torch.device([d for d in device_map.values() if d not in ["cpu", "disk"]][0])

# If shard_file is""", we use the existing state_dict instead of loading it
# If shard_file is "", we use the existing state_dict instead of loading it
if shard_file != "":
state_dict = load_state_dict(
shard_file, is_quantized=is_quantized, map_location=map_location, weights_only=weights_only
Expand Down Expand Up @@ -4895,9 +4896,10 @@ def _load_pretrained_model(
else:
model_to_load.load_state_dict(state_dict, strict=False, assign=assign_params)

# force memory release
del state_dict
gc.collect()
# force memory release if loading multiple shards
if has_multiple_shards:
gc.collect()

# Adjust offloaded weights name and save if needed
if disk_offload_index is not None and len(disk_offload_index) > 0:
Expand Down