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

Support FP16 for user buffer #690

Merged
merged 6 commits into from
Mar 7, 2024
+5 −2
Merged

Support FP16 for user buffer #690

Changes from 1 commit
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
34c1322
add_dtype_for_userbuf
Feb 29, 2024
a4c8ad9
Update transformer_engine/pytorch/module/base.py
timmoon10 Mar 6, 2024
835e5f3
Merge branch 'main' into add_dtype_for_userbuf
timmoon10 Mar 6, 2024
20298d8
Fix syntax
ksivaman Mar 6, 2024
2dec56f
Merge branch 'main' into add_dtype_for_userbuf
ksivaman Mar 6, 2024
cad2c2a
Fix lint
ksivaman Mar 7, 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
3 changes: 2 additions & 1 deletion transformer_engine/pytorch/module/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,7 @@ def initialize_ub(
shape: list,
tp_size: int,
use_fp8: bool = False,
dtype: torch.dtype = torch.bfloat16
ksivaman marked this conversation as resolved.
Show resolved Hide resolved
ub_cfgs: Optional[dict] = None
) -> None:
"""Initialize communicators for TP comm overlap using userbuffers."""
Expand Down Expand Up @@ -151,7 +152,7 @@ def add_ub(
num_splits: int = 4,
aggregate: int = 0,
) -> None:
dtype = torch.uint8 if (use_fp8 and name in fp8_buf) else torch.bfloat16
dtype = torch.uint8 if (use_fp8 and name in fp8_buf)
timmoon10 marked this conversation as resolved.
Show resolved Hide resolved
sample_buffer = torch.empty(shape, dtype=dtype, device='cuda')
if method == 'ring_exchange':
ub_obj = tex.UbufP2PCommOverlap(
Expand Down