refactor: generate_ar 改用流式输出并去除冗余注释

scripts/demo/generate_ar.py

@@ -11,7 +11,6 @@ PARAMETER_ROOT = Path(PROJECT_ROOT, "params")
 
 
 def generate_text():
-    # Load model from pretrained
     model = AutoModel.from_pretrained(PARAMETER_ROOT)
     tokenizer = AutoTokenizer.from_pretrained(PARAMETER_ROOT)
     model.to(device="cuda", dtype=torch.bfloat16)
@@ -22,16 +21,15 @@ def generate_text():
         model=model,
         tokenizer=tokenizer,
     )
-    response = engine.generate(
+    for token in engine.generate(
         prompt=query,
-        stream=False,
+        stream=True,
         max_tokens=2048,
         temperature=0.8,
         top_p=0.95,
         top_k=50,
-    )
-
-    print(response)
+    ):
+        print(token, end="", flush=True)
 
 
 if __name__ == "__main__":