Skip to content
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

Jump to bottom

make garbage collection interval configurable #533

Merged
merged 4 commits into from
Apr 2, 2024
Merged

make garbage collection interval configurable #533

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2165eda
make garbage collection interval configurable
epwalsh Apr 1, 2024
989f799
Allow disabling manual gc
epwalsh Apr 1, 2024
43f389d
Merge branch 'main' into epwalsh/gen1-gc-interval
2015aroras Apr 2, 2024
7dbce3a
Merge branch 'main' into epwalsh/gen1-gc-interval
epwalsh Apr 2, 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
3 changes: 2 additions & 1 deletion olmo/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -987,9 +987,10 @@ class TrainConfig(BaseConfig):
How often to log to the console.
"""

gen1_gc_interval: int = 1
gen1_gc_interval: Optional[int] = 1
Copy link
Collaborator

Choose a reason for hiding this comment

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

Just checking, we want the default behavior to be messing with GC?

Copy link
Member Author

Choose a reason for hiding this comment

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

Yes I think so. There might be a better default interval for smaller models, but this was a huge improvement for the 70B.

"""
How often (in steps) to run generation 1 garbage collection.
Set to ``None`` to use automatic garbage collection (i.e. we don't mess with it).
"""

compile: Optional[CompilerConfig] = None
Expand Down
5 changes: 3 additions & 2 deletions olmo/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -948,7 +948,8 @@ def fit(self):
self._gc_init_state = gc.isenabled() # cache if garbage collection is enabled, reset on close.

# Disable automatic garbage collection, FSDP doesn't work well with it.
gc.disable()
if self.cfg.gen1_gc_interval is not None:
gc.disable()

if self.cfg.load_path is not None and self.global_step > 0 and self.cfg.eval_on_load:
eval_metrics = self.eval()
Expand Down Expand Up @@ -1155,7 +1156,7 @@ def on_trace_ready(p):
break

# Run generation 1 garbage collection.
if self.global_step % self.cfg.gen1_gc_interval == 0:
if self.cfg.gen1_gc_interval is not None and self.global_step % self.cfg.gen1_gc_interval == 0:
gc.collect(1)

# Python Profiler stuff
Expand Down
Loading