Update app.py

app.py

@@ -53,6 +53,7 @@ def get_florence2():
             attn_implementation="eager",  # 關鍵：不依賴 flash_attn
             torch_dtype=torch.float16 if device == "cuda" else torch.float32
         ).to(device).eval()
+        _model.config.use_cache = False
     return _processor, _model
 
 @torch.inference_mode()
@@ -77,8 +78,9 @@ def florence2_text(image: Image.Image, task: str = "caption"):
         **inputs,
         max_new_tokens=128,
         do_sample=False,
-        num_beams=3,
-        early_stopping=True,
+        num_beams=1,
+        use_cache=False,      # ← 關掉 KV-cache（關鍵）
+        early_stopping=False, # ← 與 num_beams=1 時無效，但設 False 更乾淨
         eos_token_id=getattr(getattr(proc, "tokenizer", None), "eos_token_id", None),
     )
     out = proc.batch_decode(ids, skip_special_tokens=True)[0].strip()