Update app.py

app.py

@@ -10,8 +10,8 @@ from transformers import MoonshineForConditionalGeneration, AutoProcessor
 device = "cuda:0" if torch.cuda.is_available() else "cpu"
 torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
 
-model = MoonshineForConditionalGeneration.from_pretrained('UsefulSensors/moonshine-tiny').to(device).to(torch_dtype)
-processor = AutoProcessor.from_pretrained('UsefulSensors/moonshine-tiny')
+model = MoonshineForConditionalGeneration.from_pretrained("UsefulSensors/moonshine-tiny").to(device).to(torch_dtype)
+processor = AutoProcessor.from_pretrained("UsefulSensors/moonshine-tiny")
 
 TOKENS_PER_SEC = 12.0
 MIN_NEW_TOKENS = 48
@@ -28,13 +28,13 @@ def transcribe_audio(audio_file):
     if sr != target_sr:
         audio_array = librosa.resample(audio_array, orig_sr=sr, target_sr=target_sr)
     inputs = processor(audio_array, sampling_rate=target_sr, return_tensors="pt")
-    inputs = {k: v.to(device) for k, v in inputs.items()}
+    inputs = {k: v.to(device=device, dtype=torch_dtype) for k, v in inputs.items()}
     duration_sec = len(audio_array) / float(target_sr)
     max_new_tokens = min(MAX_NEW_TOKENS_CAP, max(MIN_NEW_TOKENS, int(math.ceil(duration_sec * TOKENS_PER_SEC))))
     generated_ids = model.generate(**inputs, do_sample=False, max_new_tokens=max_new_tokens, no_repeat_ngram_size=4, repetition_penalty=1.05)
     return processor.decode(generated_ids[0], skip_special_tokens=True)
 
-theme = gr.themes.Ocean(primary_hue="indigo", secondary_hue="fuchsia", neutral_hue="slate").set(button_large_radius='*radius_sm')
+theme = gr.themes.Ocean(primary_hue="indigo", secondary_hue="fuchsia", neutral_hue="slate").set(button_large_radius="*radius_sm")
 
 with gr.Blocks(theme=theme) as demo:
     gr.Markdown("## Moonshine Tiny STT - 27M Parameters")