Refactor app.py to streamline model setup by removing unnecessary device specification and loading the model directly in int8 format, enhancing code clarity.

app.py

@@ -20,23 +20,18 @@ torch.set_num_threads(4)
 
 logging.set_verbosity_error()
 
-# —— Model & device setup ——
-device = "cpu"
+# —— Model setup ——
 model_id = "kingabzpro/whisper-base-urdu-full"
 
-# Load in fp32 and quantize to int8
+# Load and quantize to int8
 model = AutoModelForSpeechSeq2Seq.from_pretrained(
     model_id,
-    torch_dtype=torch.float32,
     use_safetensors=True,
 )
-model.eval()
 model = torch.quantization.quantize_dynamic(model, {torch.nn.Linear}, dtype=torch.qint8)
 
 processor = AutoProcessor.from_pretrained(model_id)
 
-processor = AutoProcessor.from_pretrained(model_id)
-
 # Build a CPU-based pipeline with chunking
 transcriber = pipeline(
     task="automatic-speech-recognition",