From 4ebe9e1e2f630903a28a6c881b7a6901ea26f394 Mon Sep 17 00:00:00 2001 From: Hexq0210 <893781835@qq.com> Date: Sat, 28 Feb 2026 16:22:45 +0800 Subject: [PATCH] [NPU] bugfix: resolve modelslim load weights bug (#19472) --- .../srt/layers/quantization/modelslim/modelslim.py | 9 --------- 1 file changed, 9 deletions(-) diff --git a/python/sglang/srt/layers/quantization/modelslim/modelslim.py b/python/sglang/srt/layers/quantization/modelslim/modelslim.py index 2106d6568..1212a40d4 100644 --- a/python/sglang/srt/layers/quantization/modelslim/modelslim.py +++ b/python/sglang/srt/layers/quantization/modelslim/modelslim.py @@ -239,15 +239,6 @@ class ModelSlimConfig(QuantizationConfig): ): # adapted from vllm.model_executor.layers.quantization.utils.quant_utils.is_layer_skipped proj_name = prefix.split(".")[-1] - if not hasattr(self, "_quant_description_normalized"): - quant_description = {} - for prefix_, value in self.quant_description.items(): - prefix_ = prefix_.replace("language_model.", "") - if "visual" in prefix_: - prefix_ = prefix_.replace("model.", "") - quant_description[prefix_] = value - self.quant_description = quant_description - self._quant_description_normalized = True if proj_name in fused_mapping: shard_prefixes = [ prefix.replace(proj_name, shard_proj_name)