Skip to content

[CUTE][SM100] Fix backward gqa on sm100 post mask-mod semantic change #2146

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
drisspg merged 1 commit into from
Jan 10, 2026
+2 −2
Merged

[CUTE][SM100] Fix backward gqa on sm100 post mask-mod semantic change #2146

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
4 changes: 2 additions & 2 deletions tests/cute/test_mask_mod.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -95,7 +95,7 @@ def compute_reference_flex_attn(tensors, mask_mod_flex, block_size: tuple[int, i
device=q.device,
**block_mask_kwargs,
)
out_ref = flex_attention(q, k, v, block_mask=block_mask, scale=scale)
out_ref = flex_attention(q, k, v, block_mask=block_mask, scale=scale, enable_gqa=True)
return out_ref.transpose(1, 2).contiguous()


Expand Down Expand Up @@ -809,7 +809,7 @@ def run_flex_reference_bwd(q, k, v, block_mask, grad_out, dtype=None):

# Use flex_attention directly without torch.compile for backward tests
# torch.compile can hang on certain mask patterns (e.g., mini_causal with float32)
out_ref = flex_attention(q_ref, k_ref, v_ref, block_mask=block_mask)
out_ref = flex_attention(q_ref, k_ref, v_ref, block_mask=block_mask, enable_gqa=True)
dq_ref, dk_ref, dv_ref = torch.autograd.grad(out_ref, (q_ref, k_ref, v_ref), grad_out_ref)

# Transpose back to BSHD
Expand Down