Skip to content

[Pallas] xsa TPU sweep: use bfloat16 (was float16, hit Mosaic vector-type bug) #3204

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
norx1991 merged 1 commit into from
Jul 31, 2026
+8 −3
Merged

[Pallas] xsa TPU sweep: use bfloat16 (was float16, hit Mosaic vector-type bug) #3204

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
11 changes: 8 additions & 3 deletions benchmarks/run_tpu.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -837,9 +837,14 @@ def _xsa_shapes(
configs = configs[:num_shapes]
out: list[tuple[str, tuple[Any, ...]]] = []
for b, h, t, d in configs:
q = torch.randn(b, h, t, d, device=DEVICE, dtype=torch.float16)
k = torch.randn(b, h, t, d, device=DEVICE, dtype=torch.float16)
v = torch.randn(b, h, t, d, device=DEVICE, dtype=torch.float16)
# bfloat16, not float16: TPU Mosaic hits an "Invalid vector type for
# load" codegen error on fp16 attention shapes (the vector<8x128x2xf16>
# layout fp16 forces). bfloat16 sidesteps this; it's also the dtype
# test_examples.test_xsa uses (via HALF_DTYPE on TPU) and the sweep
# convention for every other kernel here.
q = torch.randn(b, h, t, d, device=DEVICE, dtype=torch.bfloat16)
k = torch.randn(b, h, t, d, device=DEVICE, dtype=torch.bfloat16)
v = torch.randn(b, h, t, d, device=DEVICE, dtype=torch.bfloat16)
out.append((f"[{b},{h},{t},{d}]", (q, k, v)))
return out

Expand Down
Loading