Upload app.py

app.py

@@ -4,16 +4,9 @@ from torch import exp
 from inflections_funcs import start_model, make_beams, get_beam_tokens, calculate_score_vectors
 
 # Global model initialization
-model = None
-processor = None
-
-def initialize_model():
-    global model, processor
-    if model is None:
-        print("Loading model...")
-        model, processor = start_model()
-        print("Model loaded.")
-    return model, processor
+print("Loading model...")
+model, processor = start_model()
+print("Model loaded.")
 
 def generate_beam_html(index, all_beams_data, dark_mode):
     """
@@ -70,9 +63,6 @@ def predict(prompt, dark_mode):
     """
     Generates responses for 5 beams and returns the first beam's visualization and visibility for controls.
     """
-    # Ensure model is loaded within the GPU context to avoid ZeroGPU CUDA init errors
-    model, processor = initialize_model()
-
     # Generate beams
     generated_dicts, transcription = make_beams(model, processor, prompt)