[LoRA] add support for more Qwen LoRAs (huggingface#12581)

* fix bug when offload and cache_latents both enabled

* fix

Author: linoytsaban

src/diffusers/loaders/lora_conversion_utils.py

@@ -2213,6 +2213,10 @@ def convert_key(key: str) -> str:
 
         state_dict = {convert_key(k): v for k, v in state_dict.items()}
 
+    has_default = any("default." in k for k in state_dict)
+    if has_default:
+        state_dict = {k.replace("default.", ""): v for k, v in state_dict.items()}
+
     converted_state_dict = {}
     all_keys = list(state_dict.keys())
     down_key = ".lora_down.weight"

src/diffusers/loaders/lora_pipeline.py

@@ -4940,7 +4940,8 @@ def lora_state_dict(
         has_alphas_in_sd = any(k.endswith(".alpha") for k in state_dict)
         has_lora_unet = any(k.startswith("lora_unet_") for k in state_dict)
         has_diffusion_model = any(k.startswith("diffusion_model.") for k in state_dict)
-        if has_alphas_in_sd or has_lora_unet or has_diffusion_model:
+        has_default = any("default." in k for k in state_dict)
+        if has_alphas_in_sd or has_lora_unet or has_diffusion_model or has_default:
             state_dict = _convert_non_diffusers_qwen_lora_to_diffusers(state_dict)
 
         out = (state_dict, metadata) if return_lora_metadata else state_dict