Skip to content

[fix] fix blackwell gdn accuracy issue #3156

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
kahyunnam merged 1 commit into from
Apr 24, 2026
+2 −3
Merged

[fix] fix blackwell gdn accuracy issue #3156

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
3 changes: 1 addition & 2 deletions flashinfer/gdn_kernels/blackwell/gated_delta_net_chunked.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3333,8 +3333,7 @@ def compute_group_1(

gate_handle = load_gate_consumer.wait_and_advance()

max_coord = tTR_tCcShared[cute.size(tTR_tCcShared) - 1]
cumprod_total = sCumprod[max_coord[1], 0, gate_handle.index]
cumprod_total = sCumprod[sCumprod.shape[0] - 1, 0, gate_handle.index]
Comment thread
Observer007 marked this conversation as resolved.

valid_state = not is_first_chunk or self.use_initial_state
if cutlass.const_expr(valid_state):
Expand Down
2 changes: 1 addition & 1 deletion tests/gdn/test_prefill_delta_rule.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -145,7 +145,7 @@ def _test_prefill_kernel(
atol_kv = 5e-3
rtol_kv = 1e-3
else:
atol_o = 2e-3
atol_o = 1e-3
rtol_o = 1e-3
atol_kv = 1e-3
rtol_kv = 1e-4
Expand Down
Loading