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
[Bugfix][CI] Fix config resolving logic with remote models (#27610)
Roger Wang committed
Oct 27, 2025 at 17:48 UTC
a8d2e326ecb70876aa73dce70dbe2434c64b710a
1 file changed
+8
-3
vllm/transformers_utils/config.py
+8
-3
@@ -622,9 +622,14 @@ def get_config(
622
# Architecture mapping for models without explicit architectures field
623
if not config.architectures:
624
if config.model_type not in MODEL_MAPPING_NAMES:
625
- raise ValueError(f"Cannot find architecture name for {config.model_type}")
626
- model_type = MODEL_MAPPING_NAMES[config.model_type]
627
- config.update({"architectures": [model_type]})
625
+ logger.warning(
626
+ "Model config does not have a top-level 'architectures' field: "
627
+ "expecting `hf_overrides={'architectures': ['...']}` to be passed "
628
+ "in engine args."
629
+ )
630
+ else:
631
+ model_type = MODEL_MAPPING_NAMES[config.model_type]
632
+ config.update({"architectures": [model_type]})
633
634
# ModelOpt 0.31.0 and after saves the quantization config in the model
635
# config file.