use cache

app.py

@@ -6,11 +6,18 @@ from whisper.tokenizer import get_tokenizer
 
 import classify
 
+model_cache = {}
+
 
 def zero_shot_classify(audio_path: str, class_names: str, model_name: str) -> Dict[str, float]:
     class_names = class_names.split(",")
     tokenizer = get_tokenizer(multilingual=".en" not in model_name)
-    model = whisper.load_model(model_name)
+
+    if model_name not in model_cache:
+        model = whisper.load_model(model_name)
+        model_cache[model_name] = model
+    else:
+        model = model_cache[model_name]
 
     internal_lm_average_logprobs = classify.calculate_internal_lm_average_logprobs(
         model=model,