A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Perf] Fix jit compiles at runtime of fla gated delta rule (#25432)
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Corey Lowman committed
Sep 23, 2025 at 23:16 UTC
d747c2ef18f9d9174bf40c58ae313ab31cdba305
1 file changed
+2
-2
vllm/model_executor/layers/fla/ops/fused_recurrent.py
+2
-2
@@ -40,8 +40,8 @@ def fused_recurrent_gated_delta_rule_fwd_kernel(
40
ssm_state_indices,
41
num_accepted_tokens,
42
scale,
43
- N: tl.constexpr, # num of sequences
44
- T: tl.constexpr, # num of tokens
43
+ N: tl.int64, # num of sequences
44
+ T: tl.int64, # num of tokens
45
B: tl.constexpr,
46
H: tl.constexpr,
47
HV: tl.constexpr,