Update app.py

app.py

@@ -57,6 +57,9 @@ def apply_delta(base_model_path, target_model_path, delta_path):
     base.load_state_dict(target_weights)
     # base.save_pretrained(target_model_path)
     # delta_tokenizer.save_pretrained(target_model_path)
+
+    delta = None
+    
     return base, delta_tokenizer
 
 
@@ -64,6 +67,7 @@ base_weights = 'decapoda-research/llama-7b-hf'
 target_weights = 'expertllama' # local path
 delta_weights = 'OFA-Sys/expertllama-7b-delta'
 model, tokenizer = apply_delta(base_weights, target_weights, delta_weights)
+model = model.to(torch.float)
 
 # tokenizer = transformers.LlamaTokenizer.from_pretrained(expertllama_path)
 # model = transformers.LlamaForCausalLM.from_pretrained(expertllama_path, torch_dtype=torch.float16, low_cpu_mem_usage=True)