hailin
/
jd_train
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

.

This commit is contained in:
hailin 2025-08-26 14:38:26 +08:00
parent c063c2dc06
commit fa3fce0e40
1 changed files with 1 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
train_sft_ds.py
View File

@ -301,7 +301,7 @@ def parse_args():
ap.add_argument("--save_steps", type=int, default=500) ap.add_argument("--save_steps", type=int, default=500)
ap.add_argument("--eval_ratio", type=float, default=0.0) ap.add_argument("--eval_ratio", type=float, default=0.0)
ap.add_argument("--seed", type=int, default=1337) ap.add_argument("--seed", type=int, default=1337)
ap.add_argument("--deepspeed", type=str, default="ds_config_zero3.json") #ap.add_argument("--deepspeed", type=str, default="ds_config_zero3.json")
ap.add_argument("--gradient_checkpointing", action="store_true") ap.add_argument("--gradient_checkpointing", action="store_true")
ap.add_argument("--bf16", action="store_true", ap.add_argument("--bf16", action="store_true",
help="3090/A100/H100 等可开 bf16；同时在 DS 配置里也要开") help="3090/A100/H100 等可开 bf16；同时在 DS 配置里也要开")