Skip to content

[megatron] test_convert_precision support attention fused #9395

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

Merged
Jintao-Huang merged 2 commits into from
May 21, 2026
Merged

[megatron] test_convert_precision support attention fused #9395

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
4b1d77e
test_convert_precision suppot attention_fused
Jintao-Huang May 21, 2026
5b4e755
Merge branch 'main' into test_convert_precision_suppot_attention_fused
Jintao-Huang May 21, 2026
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: 4 additions & 4 deletions swift/megatron/utils/convert_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -175,7 +175,7 @@ def broadcast_mg_logits(mg_logits=None, src_rank=None):


@contextmanager
def _patch_attention_flash_fp32(compute_dtype):
def _patch_attention_fp32(compute_dtype):
forward = TEDotProductAttention.forward

def new_forward(self, query_layer, key_layer, value_layer, *args, **kwargs):
Expand Down Expand Up @@ -252,11 +252,11 @@ def test_convert_precision(args, hf_model, mg_model, template, test_convert_dtyp
for n, m in mg_language_model.named_modules():
if n.endswith('router'):
m.to(mg_dtype)
attention_flash_context = (
_patch_attention_flash_fp32(mg_dtype) if args.attention_backend.name == 'flash' else nullcontext())
attention_context = (
_patch_attention_fp32(mg_dtype) if args.attention_backend.name in {'flash', 'fused'} else nullcontext())
Comment thread
Jintao-Huang marked this conversation as resolved.
with torch.inference_mode(), _model_cpu_forward_context(
mg_modules, test_convert_dtype, 'cuda', share_embedding=share_embedding,
target_device=mg_device), attention_flash_context:
target_device=mg_device), attention_context:
mg_logits = forward_step_helper(mg_model, mg_inputs, dtype=test_convert_dtype)
if args.tensor_model_parallel_size > 1 and args.task_type != 'seq_cls':
if mg_logits is not None:
Expand Down
Loading