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

Improve error message for speed monitor #2801

Merged
merged 3 commits into from
Jan 4, 2024
Merged

Improve error message for speed monitor #2801

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
7b5d49e
fix flops
mvpatel2000 Dec 28, 2023
512f2d2
stacklevel
mvpatel2000 Jan 2, 2024
f4b725b
Merge branch 'dev' into mvpatel2000/speed-monitor
mvpatel2000 Jan 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
19 changes: 9 additions & 10 deletions composer/callbacks/speed_monitor.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,20 +107,19 @@ def get_gpu_flops_available(state: State):
device_name = 'v100-pcie'
elif 't4' in device_name:
device_name = 't4'
else:
device_name = None

if device_name is not None:
try:
gpu_flops_available = int(GPU_AVAILABLE_FLOPS[device_name][state.precision.value])
except:
gpu_flops_available = None
if device_name in GPU_AVAILABLE_FLOPS and state.precision.value in GPU_AVAILABLE_FLOPS[device_name]:
gpu_flops_available = int(GPU_AVAILABLE_FLOPS[device_name][state.precision.value])
else:
gpu_flops_available = None

if gpu_flops_available is None:
warnings.warn(
f'gpu_flop count not found for {device_name} with precision: {state.precision.value}; ' +\
f'MFU cannot be calculated and reported. gpu_flops_available can be manually' +\
f'overridden by setting gpu_flops_available in SpeedMonitor.'
f'gpu_flop count not found for {device_name} with precision={state.precision.value} ' +\
f'so MFU cannot be calculated and reported. gpu_flops_available can be manually ' +\
f'overridden by setting gpu_flops_available in SpeedMonitor or {device_name} can ' +\
f'be added to GPU_AVAILABLE_FLOPS in composer/callbacks/speed_monitor.py',
stacklevel=2,
)
# Setting to 0 will disable MFU computation and prevent
# the speed monitor from running this helper every batch
Expand Down
Loading