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, 07:15 UTC
improve the robustness of parsing vlms config in AutoRound (#18894)
Signed-off-by: wenhuach21 <wenhua.cheng@intel.com>
Wenhua Cheng committed
May 30, 2025 at 10:24 UTC
3de3eadf5b1c271ccd7140526ffb3f850d6b0189
1 file changed
+3
-2
vllm/model_executor/layers/quantization/auto_round.py
+3
-2
@@ -116,8 +116,9 @@ class AutoRoundConfig(QuantizationConfig):
116
117
quantized = True
118
if self.block_name_to_quantize:
119
- quantized = any(name in layer_name
120
- for name in self.block_name_to_quantize)
119
+ quantized = any(
120
+ layer_name.startswith(name)
121
+ for name in self.block_name_to_quantize)
122
elif isinstance(layer, ParallelLMHead):
123
quantized = False
124