mARTin-B78/tts-dgx-spark-faster-qwen3-tts
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(customvoice): override max_seq_len to 1024 to prevent CUDA OOM

Browse Source
This commit is contained in:
mARTin-B78 2026-06-22 11:17:39 +02:00
parent 377b9f16e5
commit 79d8d2cd96
1 changed files with 2 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
config/run_customvoice_server.py
View File

@ -252,6 +252,8 @@ def main():
parser.add_argument("--device", default="cuda") parser.add_argument("--device", default="cuda")
parser.add_argument("--max-seq-len", type=int, default=2048) parser.add_argument("--max-seq-len", type=int, default=2048)
args = parser.parse_args() args = parser.parse_args()
# Force a smaller max_seq_len to save VRAM and prevent OOM
args.max_seq_len = 1024
DEFAULT_MAX_NEW_TOKENS = args.max_seq_len DEFAULT_MAX_NEW_TOKENS = args.max_seq_len
_load_model_kwargs = args _load_model_kwargs = args