Update app.py

app.py

@@ -92,16 +92,20 @@ def load_model(model_path, progress=gr.Progress()):
     
     try:
         progress(0.3, desc="Loading tokenizer...")
-        config = AutoConfig.from_pretrained(model_path, trust_remote_code=True,force_download=True)
-        current_tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False,trust_remote_code=True,force_download=True)
+        config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
+        if 'MoE' in model_path:
+            config_moe = config
+            config_moe["auto_map"]["AutoConfig"] = "./configuration_upcycling_qwen2_moe.UpcyclingQwen2MoeConfig"
+            config_moe["auto_map"]["AutoModelForCausalLM"] = "./modeling_upcycling_qwen2_moe.UpcyclingQwen2MoeForCausalLM"
+        current_tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False,trust_remote_code=True)
         
         progress(0.5, desc="Loading model...")
         current_model = AutoModelForCausalLM.from_pretrained(
             model_path, 
             device_map="auto", 
             torch_dtype=torch.float16,
-            config=config,
-            trust_remote_code=True,force_download=True
+            config=config_moe if 'MoE' in model_path else config,
+            trust_remote_code=True
         )
         
         current_model_path = model_path