Skip to content

Log GPU memory metrics during training #8

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
Creampelt wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Log GPU memory metrics during training #8

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
14 changes: 14 additions & 0 deletions robot_rl/runners/on_policy_runner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -238,6 +238,20 @@ def log(self, locs: dict, width: int = 80, pad: int = 35):
self.writer.add_scalar("Perf/total_fps", fps, locs["it"])
self.writer.add_scalar("Perf/collection time", locs["collection_time"], locs["it"])
self.writer.add_scalar("Perf/learning_time", locs["learn_time"], locs["it"])
# -- GPU Performance
for gpu_i in range(self.gpu_world_size):
mem_allocated = torch.cuda.memory_allocated(gpu_i)
mem_reserved = torch.cuda.memory_reserved(gpu_i)
total_mem = torch.cuda.get_device_properties(gpu_i).total_memory
max_mem_allocated = torch.cuda.max_memory_allocated(gpu_i)
max_mem_reserved = torch.cuda.max_memory_reserved(gpu_i)

self.writer.add_scalar(f"Perf/gpu.{gpu_i}/mem_allocated_GB", mem_allocated / 1e9, locs["it"])
self.writer.add_scalar(f"Perf/gpu.{gpu_i}/mem_reserved_GB", mem_reserved / 1e9, locs["it"])
self.writer.add_scalar(f"Perf/gpu.{gpu_i}/mem_allocated_pct", mem_allocated / total_mem, locs["it"])
self.writer.add_scalar(f"Perf/gpu.{gpu_i}/mem_reserved_pct", mem_reserved / total_mem, locs["it"])
self.writer.add_scalar(f"Perf/gpu.{gpu_i}/max_mem_allocated_GB", max_mem_allocated / 1e9, locs["it"])
self.writer.add_scalar(f"Perf/gpu.{gpu_i}/max_mem_reserved_GB", max_mem_reserved / 1e9, locs["it"])

# -- Training
if len(locs["rewbuffer"]) > 0:
Expand Down