Update generate.py

generate.py

@@ -21,17 +21,17 @@ TYPE_WRITER=1 # whether output streamly
 
 args = parser.parse_args()
 print(args)
-tokenizer = LlamaTokenizer.from_pretrained(args.model_path)
+tokenizer = LlamaTokenizer.from_pretrained(BASE_MODE)
 
 LOAD_8BIT = True
 
 
 
 # fix the path for local checkpoint
-lora_bin_path = os.path.join(args.lora_path, "adapter_model.bin")
+lora_bin_path = os.path.join(LORA_PATH, "adapter_model.bin")
 print(lora_bin_path)
-if not os.path.exists(lora_bin_path) and args.use_local:
-    pytorch_bin_path = os.path.join(args.lora_path, "pytorch_model.bin")
+if not os.path.exists(lora_bin_path) and USE_LOCAL:
+    pytorch_bin_path = os.path.join(LORA_PATH, "pytorch_model.bin")
     print(pytorch_bin_path)
     if os.path.exists(pytorch_bin_path):
         os.rename(pytorch_bin_path, lora_bin_path)
@@ -140,7 +140,7 @@ def evaluate(
         **kwargs,
     )
     with torch.no_grad():
-        if args.use_typewriter:
+        if TYPE_WRITER:
             for generation_output in model.stream_generate(
                 input_ids=input_ids,
                 generation_config=generation_config,