Improve flops profiler functionality #1065
cli99
merged
22 commits into
deepspeedai:master
from
cli99:fix-flops-profiler
use the original function's name as the key to old_functions dict
a84adf9f
add option to write flops profiler output to a file
f1219b27
update profile output format
172c4ab9
print at glocal rank 0
bd47aee4
add flops calculation in bwd pass using time from ds timers
154e791f
improve aggregated profiling out to show all depth
855b5e2d
print samples/second
b8e03aaa
use the original function's name as the key to old_functions dict
bab474f6
add option to write flops profiler output to a file
fd48a271
update profile output format
326eff84
print at glocal rank 0
f8c613c6
add flops calculation in bwd pass using time from ds timers
34b5b9f4
improve aggregated profiling out to show all depth
136f61d7
print samples/second
182292c8
Merge branch 'fix-flops-profiler' of https://github.com/cli99/DeepSpe…
1f98d371
Merge branch 'master' into fix-flops-profiler
cda7d594
update readme and examples
8fd2cf3d
Merge branch 'fix-flops-profiler' of https://github.com/cli99/DeepSpe…
875fcd7b
update docs
3d646066
tjruwase
approved these changes
on 2021-05-12
tjruwase
commented
on 2021-05-12
tjruwase
commented
on 2021-05-12
fix typo and reorder printing
cac9a36e
tjruwase
approved these changes
on 2021-05-12
fix format
b8038b6a
fix conflicts
5a9796c6
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub