Update stf/stf-api-alternative/src/stf_alternative/inference.py

stf/stf-api-alternative/src/stf_alternative/inference.py

@@ -141,7 +141,7 @@ def process_audio_chunk(audio_processor, audio_encoder, audio_chunk, device):
 
     input_values = audio_processor(
         audio_data, sampling_rate=16000, return_tensors="pt"
-    ).cuda(0)["input_values"] #//.to(device)["input_values"]
+    ).to(device)["input_values"]
 
     #with torch.no_grad():
     logits = audio_encoder(input_values=input_values)