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 5, 09:45 UTC
[UX] Improve error message for MM input too long (#39409)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Apr 9, 2026 at 21:20 UTC
54d244f28f57b5ca2e6afdb91786b5c798adb42f
1 file changed
+3
-3
vllm/v1/engine/input_processor.py
+3
-3
@@ -405,11 +405,11 @@ class InputProcessor:
405
decoder_mm_positions = prompt_input["mm_placeholders"]
406
for modality, mm_positions in decoder_mm_positions.items():
407
for mm_position in mm_positions:
408
- embed_length = mm_position.get_num_embeds()
409
- if embed_length > self.mm_encoder_cache_size:
408
+ num_embeds = mm_position.get_num_embeds()
409
+ if num_embeds > self.mm_encoder_cache_size:
410
raise ValueError(
411
f"The {prompt_type} prompt contains a(n) {modality} item "
412
- f"with length {embed_length}, which exceeds the "
412
+ f"with {num_embeds} embedding tokens, which exceeds the "
413
f"pre-allocated encoder cache size "
414
f"{self.mm_encoder_cache_size}. Please reduce the input "
415
f"size or increase the encoder cache size "