up

run_demo_openai.py

@@ -62,7 +62,9 @@ def maybe_load_cached_pipeline(model_name):
     if model is None:
         downloaded_model_path = hf_hub_download(repo_id=model_name, filename=CHECKPOINT_FILENAME)
 
-        model = whisper.load_model(downloaded_model_path, device=device)
+        # model = whisper.load_model(downloaded_model_path, device=device)
+        model = whisper.load_model(downloaded_model_path, device="cpu")
+        model = model.to(device)
         logger.info(f"`{model_name}` has been loaded on device `{device}`")
 
         print_memory_info()
@@ -101,7 +103,7 @@ def transcribe(microphone, file_upload, with_timestamps, model_name=DEFAULT_MODE
     # text = model.transcribe(file, **GEN_KWARGS)["text"]
     text = infer(model, file, with_timestamps)
 
-    logger.info(f"Transcription by `{model_name}`: {text}")
+    logger.info(f"Transcription by `{model_name}`:\n{text}\n")
 
     return warn_output + text
 
@@ -124,7 +126,7 @@ def yt_transcribe(yt_url, with_timestamps, model_name=DEFAULT_MODEL_NAME):
     # text = model.transcribe("audio.mp3", **GEN_KWARGS)["text"]
     text = infer(model, "audio.mp3", with_timestamps)
 
-    logger.info(f'Transcription by `{model_name}` of "{yt_url}": {text}')
+    logger.info(f'Transcription by `{model_name}` of "{yt_url}":\n{text}\n')
 
     return html_embed_str, text