Switch to auto compute type

app.py

@@ -519,7 +519,7 @@ if __name__ == '__main__':
                         help="directory to save the outputs")
     parser.add_argument("--whisper_implementation", type=str, default=default_whisper_implementation, choices=["whisper", "faster-whisper"],\
                         help="the Whisper implementation to use")
-    parser.add_argument("--compute_type", type=str, default=default_app_config.compute_type, choices=["int8", "int8_float16", "int16", "float16"], \
+    parser.add_argument("--compute_type", type=str, default=default_app_config.compute_type, choices=["default", "auto", "int8", "int8_float16", "int16", "float16", "float32"], \
                         help="the compute type to use for inference")
 
     args = parser.parse_args().__dict__

cli.py

@@ -80,7 +80,7 @@ def cli():
                         help="if True, provide the previous output of the model as a prompt for the next window; disabling may make the text inconsistent across windows, but the model becomes less prone to getting stuck in a failure loop")
     parser.add_argument("--fp16", type=str2bool, default=app_config.fp16, \
                         help="whether to perform inference in fp16; True by default")
-    parser.add_argument("--compute_type", type=str, default=app_config.compute_type, choices=["int8", "int8_float16", "int16", "float16"], \
+    parser.add_argument("--compute_type", type=str, default=app_config.compute_type, choices=["default", "auto", "int8", "int8_float16", "int16", "float16", "float32"], \
                         help="the compute type to use for inference")
 
     parser.add_argument("--temperature_increment_on_fallback", type=optional_float, default=app_config.temperature_increment_on_fallback, \

config.json5

@@ -118,7 +118,7 @@
     // Whether to perform inference in fp16; True by default
     "fp16": true,
     // The compute type used by faster-whisper. Can be "int8". "int16" or "float16".
-    "compute_type": "float16",
+    "compute_type": "auto",
     // Temperature to increase when falling back when the decoding fails to meet either of the thresholds below
     "temperature_increment_on_fallback": 0.2,
     // If the gzip compression ratio is higher than this value, treat the decoding as failed