From 02d654760fd7116554b8cb6cf3ec0c2569330928 Mon Sep 17 00:00:00 2001 From: Renzheng Wang Date: Sun, 5 Jul 2026 17:50:43 +0800 Subject: [PATCH] [BugFix] Drop removed get_cache_scale call in HunyuanImage3 diffusion loader (#4892) Signed-off-by: wangrzneu --- .../models/hunyuan_image3/hunyuan_image3_transformer.py | 9 ++------- 1 file changed, 2 insertions(+), 7 deletions(-) diff --git a/vllm_omni/diffusion/models/hunyuan_image3/hunyuan_image3_transformer.py b/vllm_omni/diffusion/models/hunyuan_image3/hunyuan_image3_transformer.py index 1212497e6..0416fdcf1 100644 --- a/vllm_omni/diffusion/models/hunyuan_image3/hunyuan_image3_transformer.py +++ b/vllm_omni/diffusion/models/hunyuan_image3/hunyuan_image3_transformer.py @@ -2235,13 +2235,8 @@ class HunyuanImage3Model(nn.Module): # processed with quantization, LoRA, fine-tuning, etc. if self.config.tie_word_embeddings and "lm_head.weight" in name: continue - if self.quant_config is not None and (scale_name := self.quant_config.get_cache_scale(name)): - # Loading kv cache scales for compressed-tensors quantization - param = params_dict[scale_name] - weight_loader = getattr(param, "weight_loader", default_weight_loader) - loaded_weight = loaded_weight[0] - weight_loader(param, loaded_weight) - continue + # KV-cache scales are renamed via maybe_remap_kv_scale_name below; + # quant_config.get_cache_scale was removed in vLLM v0.23.0 (see #4810). is_found = False for param_name, weight_name, shard_id in stacked_params_mapping: