Skip to content

[ExecuTorch][WebGPU] et_vk.sdpa channel-attention tests #20872

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
JCNTH wants to merge 1 commit into
base: gh/JCNTH/47/base
Choose a base branch
from
+5 −1
Open

[ExecuTorch][WebGPU] et_vk.sdpa channel-attention tests #20872

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
6 changes: 5 additions & 1 deletion backends/webgpu/test/op_tests/cases.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,3 @@
# Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
Expand Down Expand Up @@ -466,7 +466,10 @@
def _et_vk_sdpa_suite() -> WebGPUTestSuite:
# Non-causal fused attention (Florence-2 vision + BART, via the et_vk source
# transform). Covers self-attn, an asymmetric S_q != S_kv (cross-attn) case,
# an additive mask (BART), and D=128 (Voxtral/DaViT) through the vec4 kernels.
# an additive mask (BART), and D=128 (Voxtral/DaViT). The QK dispatch is
# occupancy-routed: chattn_davit (num_rows = B*H*S_q = 256 < the 4096 floor)
# exercises the per-entry QK kernel; selfattn_siglip (num_rows = 6912) is the
# per-row guard. Both branches must match the same fp32 golden.
def qkv(b, h, sq, skv, d):
return (
InputSpec(shape=(b, h, sq, d), gen=_sdpa_randn),
Expand All @@ -486,6 +489,7 @@
inputs=qkv(1, 4, 8, 8, 16),
),
Case(name="d128_voxtral", inputs=qkv(1, 4, 6, 6, 128)),
Case(name="chattn_davit", inputs=qkv(1, 8, 32, 256, 64)),
],
golden_dtype="float32",
atol=1e-4,
Expand Down
Loading