Update app.py

app.py

@@ -16,14 +16,16 @@ language = "zh"
 pipe = pipeline(
     task="automatic-speech-recognition",
     model=MODEL_NAME,
-    chunk_length_s=30,
+    chunk_length_s=30,  # Reduce chunk size for better memory handling
     device=device,
     generate_kwargs={
         "no_repeat_ngram_size": 3,
-        "repetition_penalty": 1.3,
+        "repetition_penalty": 1.15,
         "temperature": 0.7,
-        "top_p": 0.9,
-        "top_k": 50
+        "top_p": 0.97,
+        "top_k": 40,
+        "max_new_tokens": 300,  # Reduced from 500 to avoid exceeding 448
+        "do_sample": True  # Required for `top_p` and `top_k` to take effect
     }
 )
 pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=language, task="transcribe")