separate endpoints

app.py

@@ -313,8 +313,10 @@ def generate_caption_text(image: Image.Image, prompt: str = CAPTION_PROMPT) -> s
             outputs = runtime_model.generate(
                 **inputs,
                 max_new_tokens=MAX_NEW_TOKENS,
-                do_sample=False,
-                num_beams=1,
+                do_sample=True,
+                top_p=0.9,
+                temperature=0.7,
+                repetition_penalty=1.2,
             )
     except Exception as exc:
         raise AppError("Caption generation failed.", 500) from exc