forked from mrq/tortoise-tts
Expose batch size and device settings in CLI
This commit is contained in:
parent
b876a6b32c
commit
fc4a31028a
|
@ -79,6 +79,12 @@ advanced_group.add_argument(
|
||||||
help='Normally text enclosed in brackets are automatically redacted from the spoken output '
|
help='Normally text enclosed in brackets are automatically redacted from the spoken output '
|
||||||
'(but are still rendered by the model), this can be used for prompt engineering. '
|
'(but are still rendered by the model), this can be used for prompt engineering. '
|
||||||
'Set this to disable this behavior.')
|
'Set this to disable this behavior.')
|
||||||
|
advanced_group.add_argument(
|
||||||
|
'--device', type=str, default=None,
|
||||||
|
help='Device to use for inference.')
|
||||||
|
advanced_group.add_argument(
|
||||||
|
'--batch-size', type=int, default=None,
|
||||||
|
help='Batch size to use for inference. If omitted, the batch size is set based on available GPU memory.')
|
||||||
|
|
||||||
tuning_group = parser.add_argument_group('tuning options (overrides preset settings)')
|
tuning_group = parser.add_argument_group('tuning options (overrides preset settings)')
|
||||||
tuning_group.add_argument(
|
tuning_group.add_argument(
|
||||||
|
@ -200,7 +206,8 @@ if args.play:
|
||||||
seed = int(time.time()) if args.seed is None else args.seed
|
seed = int(time.time()) if args.seed is None else args.seed
|
||||||
if not args.quiet:
|
if not args.quiet:
|
||||||
print('Loading tts...')
|
print('Loading tts...')
|
||||||
tts = TextToSpeech(models_dir=args.models_dir, enable_redaction=not args.disable_redaction)
|
tts = TextToSpeech(models_dir=args.models_dir, enable_redaction=not args.disable_redaction,
|
||||||
|
device=args.device, autoregressive_batch_size=args.batch_size)
|
||||||
gen_settings = {
|
gen_settings = {
|
||||||
'use_deterministic_seed': seed,
|
'use_deterministic_seed': seed,
|
||||||
'varbose': not args.quiet,
|
'varbose': not args.quiet,
|
||||||
|
|
Loading…
Reference in New Issue
Block a user