Skip to content

manual gc in llama3 #1381

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

Draft
pstjohn wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

manual gc in llama3 #1381

Show file tree
Hide file tree
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
1 change: 1 addition & 0 deletions bionemo-recipes/recipes/llama3_native_te/hydra_config/defaults.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,6 +74,7 @@ logger:

profiler:
enabled: false
gc_interval: 1_000 # Run garbage collection every 1000 steps
schedule:
wait: 10
warmup: 10
Expand Down
14 changes: 14 additions & 0 deletions bionemo-recipes/recipes/llama3_native_te/perf_logger.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@
# See the License for the specific language governing permissions and
# limitations under the License.

import gc
import logging
import time
from pathlib import Path
Expand Down Expand Up @@ -71,6 +72,11 @@ def __init__(self, dist_config: DistributedConfig, args: DictConfig):
self.previous_step_time = time.perf_counter()
self._profiler = None

# Manually control garbage collection for cleaner profiling.
self._gc_interval = args.profiler.gc_interval
gc.disable()
self._run_garbage_collection()

if self._dist_config.is_main_process():
# Log the entire args object to wandb for experiment tracking and reproducibility.
self._wandb_run = wandb.init(**args.wandb, config=self._run_config)
Expand Down Expand Up @@ -134,6 +140,9 @@ def log_step(
if self._dist_config.local_rank == 0:
logger.info(", ".join([f"{k.split('/')[1]}: {v:.3g}" for k, v in metrics.items()]))

if (step + 1) % self._gc_interval == 0:
self._run_garbage_collection()

def finish(self):
"""Finish the logger and close the progress bar."""
if self._profiler is not None:
Expand All @@ -145,6 +154,11 @@ def finish(self):
wandb.finish()
self._progress_bar.close()

def _run_garbage_collection(self):
"""Run garbage collection."""
gc.collect()
torch.cuda.empty_cache()


def setup_profiler(args: DictConfig, wandb_run: wandb.Run):
"""Setup a basic torch profiler for the experiment.
Expand Down
4 changes: 0 additions & 4 deletions bionemo-recipes/recipes/llama3_native_te/train_fsdp2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,6 @@
# See the License for the specific language governing permissions and
# limitations under the License.

import gc
import logging
from contextlib import nullcontext
from pathlib import Path
Expand Down Expand Up @@ -128,9 +127,6 @@ def main(args: DictConfig) -> float | None:

perf_logger = PerfLogger(dist_config, args)

gc.collect()
torch.cuda.empty_cache()

# Training loop
logger.info(f"Starting training loop from step {start_step} to {args.num_train_steps}")
step = start_step
Expand Down