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 6, 23:00 UTC
[Minor] Enhance error message for TRTLLM decode uniformity check (#36609)
Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
Woosuk Kwon committed
Mar 10, 2026 at 15:38 UTC
195d1ca3e8b1662e5df88b159a4306c48e1b0b5c
1 file changed
+2
-1
vllm/v1/attention/backends/flashinfer.py
+2
-1
@@ -1110,7 +1110,8 @@ class FlashInferMetadataBuilder(AttentionMetadataBuilder[FlashInferMetadata]):
1110
if num_decodes > 0:
1111
if decode_use_trtllm:
1112
assert num_decode_tokens % num_decodes == 0, (
1113
- "TRTLLM decode requires uniform query lengths per request."
1113
+ "TRTLLM decode requires uniform query lengths per request. "
1114
+ f"Got {num_decode_tokens=} and {num_decodes=}."
1115
)
1116
attn_metadata.decode = TRTLLMDecode(
1117
block_tables=block_table_tensor[:num_decodes],