if conversion is needed, we need to use the converted model state_dic…

…t(). (ModelCloud#54)

Co-authored-by: LRL-ModelCloud <[email protected]>

gptqmodel/models/base.py

@@ -500,8 +500,6 @@ def save_quantized(
                     f"gptq_model-{self.quantize_config.bits}bit-{self.quantize_config.group_size}g"
             )
 
-        state_dict = self.model.state_dict()
-
         if format == FORMAT.GPTQ_V2 or (format is None and quantize_config.format == FORMAT.GPTQ_V2):
             logger.warning(
                 f"Using 'format = {FORMAT.GPTQ_V2}': the serialized model is only supported by GPTQModel version >= {MIN_VERSION_WITH_V2}."
@@ -566,6 +564,8 @@ def save_quantized(
 
         model.to(CPU)
 
+        state_dict = model.state_dict()
+
         if quantize_config.model_file_base_name is None:
             if use_safetensors:
                 model_base_name = "model"