Skip to content

GitLab

Commit 19730725 authored by Lawrence McAfee's avatar Lawrence McAfee
Browse files

using smaller chunk_numel_min.

parent ed90a1b2
Hide whitespace changes
Inline Side-by-side
Showing with 4 additions and 1 deletion
megatron/optimizer/distrib_optimizer.py
View file @ 19730725
...@@ -358,7 +358,10 @@ class DistributedOptimizer(MixedPrecisionOptimizer): ...@@ -358,7 +358,10 @@ class DistributedOptimizer(MixedPrecisionOptimizer):
for view in gbuf_views: for view in gbuf_views:
assert view.nelement() == view_numel assert view.nelement() == view_numel
chunk_numel_min = 1024**2 # chunk_numel_min = 1024
# chunk_numel_min = 16384
chunk_numel_min = 131072
# chunk_numel_min = 1048576
chunk_numel_max = view_numel chunk_numel_max = view_numel
# chunk_numel_min_log = math.log(chunk_numel_min) # chunk_numel_min_log = math.log(chunk_numel_min)
# chunk_numel_max_log = math.log(chunk_numel_max) # chunk_numel_max_log = math.log(chunk_numel_max)
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment