[Perf] Fix jit compiles at runtime of fla gated delta rule (#25432)

Co-authored-by: Michael Goin <mgoin64@gmail.com> Signed-off-by: yewentao256 <zhyanwentao@126.com>
2025-09-23 23:16:13 -04:00
parent 177c37e960
commit f52b991db6
1 changed files with 2 additions and 2 deletions
--- a/vllm/model_executor/layers/fla/ops/fused_recurrent.py
+++ b/vllm/model_executor/layers/fla/ops/fused_recurrent.py
@ -40,8 +40,8 @@ def fused_recurrent_gated_delta_rule_fwd_kernel(
    ssm_state_indices,
    num_accepted_tokens,
    scale,
-    N: tl.constexpr,  # num of sequences
-    T: tl.constexpr,  # num of tokens
+    N: tl.int64,  # num of sequences
+    T: tl.int64,  # num of tokens
    B: tl.constexpr,
    H: tl.constexpr,
    HV: tl.constexpr,