[Bugfix][Triton] Make fp8_min/fp8_max constexpr in _quantize_pad_fp8_kernel (#51872)
Signed-off-by: Max Hu <maxhu@nvidia.com> Co-authored-by: Max Hu <maxhu@nvidia.com>
Max Hu committed
Aug 11, 2026 at 23:09 UTC
d462dee37da2e1a744c6694cb65a1c735600f262
1 file changed
+2
-2
vllm/kernels/triton/qkv_padded_fp8_quant.py
+2
-2
@@ -34,8 +34,8 @@ def _quantize_pad_fp8_kernel(
34
n_rows,
35
n_cols,
36
n_cols_padded,
37
- fp8_min,
38
- fp8_max,
37
+ fp8_min: tl.constexpr,
38
+ fp8_max: tl.constexpr,
39
SKIP_SCALE: tl.constexpr,
40
BLOCK_M: tl.constexpr,
41
BLOCK_N: tl.constexpr,