Files
datawhalechina--self-llm/models/MiniCPM/train.sh
T
2026-07-13 12:59:13 +08:00

12 lines
335 B
Bash

num_gpus=1
deepspeed --num_gpus $num_gpus train.py \
--deepspeed ./ds_config.json \
--output_dir="./output/MiniCPM" \
--per_device_train_batch_size=4 \
--gradient_accumulation_steps=4 \
--logging_steps=10 \
--num_train_epochs=3 \
--save_steps=500 \
--learning_rate=1e-4 \
--save_on_each_node=True \