Files
pretrain_kaiyuan2b/scripts/kaiyuan2b-profiling/params/qwen3_1p7b/hparams.sh
2026-05-06 15:06:07 +08:00

14 lines
247 B
Bash

HPARAMS="
--micro-batch-size 16
--global-batch-size 2048
--train-iters 19760
--eval-iters 10
--eval-interval 1000
--save-interval 1000
--log-interval 1
--lr 5e-3
--min-lr 5e-3
--lr-decay-style constant
--lr-warmup-iters 10
"