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 15, 02:15 UTC
Fix overflow indexing in causal_conv1d kernel (#20938)
Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>
Thomas Parnell committed
Jul 14, 2025 at 23:43 UTC
86f3ac21cefb51d276d94666a5168823fa0565de
1 file changed
+2
-1
vllm/model_executor/layers/mamba/ops/causal_conv1d.py
+2
-1
@@ -92,7 +92,8 @@ def _causal_conv1d_fwd_kernel( # continuous batching
92
93
if IS_CONTINUOUS_BATCHING:
94
# cache_idx
95
- conv_state_batch_coord = tl.load(conv_state_indices_ptr + idx_seq)
95
+ conv_state_batch_coord = tl.load(conv_state_indices_ptr + idx_seq).to(
96
+ tl.int64)
97
else:
98
# cache_idx
99
conv_state_batch_coord = idx_seq