[Bugfix][Spec Decode] Fix NaN handling in rejection sampler tl.argmax (#50183)

Signed-off-by: Gabriel Peracio <gabriel@sigsonic.com>
Co-authored-by: Kimi K3 <noreply@moonshot.ai>
This commit is contained in:
gabriel-peracio
2026-08-05 18:01:09 -07:00
committed by GitHub
co-authored by Kimi K3
parent d36f24b66a
commit 47a4e410ba
2 changed files with 51 additions and 0 deletions
@@ -108,6 +108,8 @@ def _compute_global_target_argmax(
mask=blocks_mask,
other=float("-inf"),
)
# See _insert_resampled_kernel: NaN breaks tl.argmax index bounds.
local_max = tl.where(local_max != local_max, float("-inf"), local_max)
max_block_idx = tl.argmax(local_max, axis=0)
return tl.load(
target_local_argmax_ptr + logit_idx * target_local_argmax_stride + max_block_idx
@@ -849,6 +851,14 @@ def _insert_resampled_kernel(
mask=mask,
other=float("-inf"),
)
# NaN max values (from NaN target logits) make tl.argmax return an
# out-of-range block index (into the padded region), causing an OOB read
# of resampled_local_argmax. Map NaN to -inf so argmax stays in range.
resampled_local_max = tl.where(
resampled_local_max != resampled_local_max,
float("-inf"),
resampled_local_max,
)
resampled_max_block_idx = tl.argmax(resampled_local_max, axis=0)
resampled = tl.load(
resampled_local_argmax_ptr