Skip to content

Add assertion for mxfp8 params without dp overlap #2271

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
wants to merge 1 commit into
base: main
Choose a base branch
from
+16 −0
Open

Add assertion for mxfp8 params without dp overlap #2271

Show file tree
Hide file tree
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
8 changes: 8 additions & 0 deletions megatron/core/distributed/distributed_data_parallel_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,6 +146,14 @@ def __post_init__(self):
"""Check the validity of the config."""
if self.reuse_grad_buf_for_mxfp8_param_ag:
assert self.fp8_param_gather, "Reuse grad buffer only when keeping params in MXFP8."
# Using mxfp8 param without overlap param gather and overlap grad reduce will cause NaN.
# TODO: Remove this assertion when the issue is fixed.
assert (
self.overlap_param_gather
), "--overlap-param-gather is required when using mxfp8 params"
assert (
self.overlap_grad_reduce
), "--overlap-grad-reduce is required when using mxfp8 params"

if self.nccl_ub:
if 'expandable_segments:True' in os.getenv('PYTORCH_CUDA_ALLOC_CONF', '').split(','):
Expand Down
8 changes: 8 additions & 0 deletions megatron/core/distributed/fsdp/src/megatron_fsdp/distributed_data_parallel_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -137,6 +137,14 @@ def __post_init__(self):
"""Check the validity of the config."""
if self.reuse_grad_buf_for_mxfp8_param_ag:
assert self.fp8_param_gather, "Reuse grad buffer only when keeping params in MXFP8."
# Using mxfp8 param without overlap param gather and overlap grad reduce will cause NaN.
# TODO: Remove this assertion when the issue is fixed.
assert (
self.overlap_param_gather
), "--overlap-param-gather is required when using mxfp8 params"
assert (
self.overlap_grad_reduce
), "--overlap-grad-reduce is required when using mxfp8 params"

if self.nccl_ub:
if 'expandable_segments:True' in os.getenv('PYTORCH_CUDA_ALLOC_CONF', '').split(','):
Expand Down