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, 14:45 UTC
Improve error for structured output backend selection (#16717)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Apr 17, 2025 at 01:35 UTC
93e561ec4dfadba837e47f63fb9d9ec971b6ba65
1 file changed
+7
-4
vllm/v1/engine/processor.py
+7
-4
@@ -155,10 +155,13 @@ class Processor:
155
"supported in V1.")
156
if params.guided_decoding.backend:
157
if params.guided_decoding.backend != engine_level_backend:
158
- raise ValueError("Request-level structured output backend "
159
- "must match engine-level backend. "
160
- f"{params.guided_decoding.backend}"
161
- f" != {engine_level_backend}")
158
+ raise ValueError(
159
+ "Request-level structured output backend selection is no "
160
+ "longer supported. The request specified "
161
+ f"'{params.guided_decoding.backend}', but vLLM was "
162
+ f"initialised with '{engine_level_backend}'. This error "
163
+ "can be resolved by removing backend selection from the "
164
+ "request.")
165
else:
166
params.guided_decoding.backend = engine_level_backend
167