This commit is contained in:
hailin 2025-09-26 15:25:16 +08:00
parent a53178dacb
commit 04b6a5c3a8
1 changed files with 2 additions and 2 deletions

View File

@ -2,8 +2,8 @@
set -euo pipefail set -euo pipefail
# ===== 可调参数 ===== # ===== 可调参数 =====
CKPT_ROOT="/home/test/checkpoints/q3-32b-lora" # 若实际是 .../checkpoint-62/global_step62请把 CKPT_ROOT 改成 .../checkpoint-62 CKPT_ROOT="/home/test/checkpoints/q3-32b-ds4/checkpoint-60" # 若实际是 .../checkpoint-62/global_step62请把 CKPT_ROOT 改成 .../checkpoint-62
TAG="global_step110" TAG="global_step60"
HOSTS=(tn01 tn02 tn03 tn04 tn05 tn06) HOSTS=(tn01 tn02 tn03 tn04 tn05 tn06)
AGGREGATOR_HOST="tn06" # 本脚本运行/汇总所在机器 AGGREGATOR_HOST="tn06" # 本脚本运行/汇总所在机器
EXPECTED_SHARDS_PER_HOST=4 # 每机应写出分片数(按你的并行布局) EXPECTED_SHARDS_PER_HOST=4 # 每机应写出分片数(按你的并行布局)