Mirrors/VoiceCraft
1
0
Fork 0
mirror of https://github.com/jasonppy/VoiceCraft.git synced 2025-12-16 11:57:44 +01:00
Code Issues Packages Projects Releases Wiki Activity

change default inference from top-p to top-k sampling, massive performance gain

Browse Source
This commit is contained in:
pyp_l40
2025-03-15 18:16:27 -05:00
parent 013a21c70d
commit 7121981bb4
6 changed files with 12 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
inference_tts_scale.py
View File

@@ -25,8 +25,8 @@ def get_args():
parser.add_argument("--seed", type=int, default=1)
parser.add_argument("--codec_audio_sr", type=int, default=16000, help='the sample rate of audio that the codec is trained for')
parser.add_argument("--codec_sr", type=int, default=50, help='the sample rate of the codec codes')
parser.add_argument("--top_k", type=int, default=0, help="sampling param")
parser.add_argument("--top_p", type=float, default=0.8, help="sampling param")
parser.add_argument("--top_k", type=int, default=40, help="sampling param")
parser.add_argument("--top_p", type=float, default=1, help="sampling param")
parser.add_argument("--temperature", type=float, default=1.0, help="sampling param")
parser.add_argument("--output_dir", type=str, default=None)
parser.add_argument("--device", type=str, default="cuda")