i forgor

app.py

@@ -18,7 +18,7 @@ processor = AutoProcessor.from_pretrained(MODEL_ID)
 def tag_image(image, max_new_tokens=128, temperature=1, top_p=1, min_p=0):
     inputs = processor(images=image, text="<image>tag en", return_tensors="pt").to("cuda")
     streamer = TextIteratorStreamer(tokenizer=processor.tokenizer, skip_prompt=True)
-    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=max_new_tokens, use_cache=True, cache_implementation="hybrid", temperature=temperature, top_p=top_p, min_p=min_p)
+    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=max_new_tokens, use_cache=True, cache_implementation="hybrid", do_sample=True, temperature=temperature, top_p=top_p, min_p=min_p)
     thread = Thread(target=model.generate, kwargs=generation_kwargs)
     
     text = ""