forked from jiuyuan/CPM-9G-8B
57 lines
1.6 KiB
Bash
57 lines
1.6 KiB
Bash
#! /bin/bash
|
|
|
|
export MASTER_ADDR="localhost"
|
|
export MASTER_PORT=12370
|
|
|
|
CPM_PATH="/workspace/repo/CPM-9G-8B/9G-Train"
|
|
|
|
NO=1
|
|
MAX_STEP=1000
|
|
GPU_NUM=multi
|
|
EXP_PATH=/workspace/repo/CPM-9G-8B/results/lora/${GPU_NUM}/${MAX_STEP}/${NO}/models/
|
|
MODEL_NAME="9g-sft"
|
|
TB_PATH="/workspace/repo/CPM-9G-8B/results/lora/${GPU_NUM}/${MAX_STEP}/${NO}/logs/"
|
|
|
|
OPTS=""
|
|
OPTS+=" --vocab /v2/sft_8b_v2/vocab.txt"
|
|
OPTS+=" --model-config /v2/sft_8b_v2/config.json"
|
|
|
|
OPTS+=" --train-iters ${MAX_STEP}"
|
|
OPTS+=" --inspect-iters 500"
|
|
OPTS+=" --warmup-iters 20"
|
|
|
|
OPTS+=" --lr-decay-style cosine"
|
|
OPTS+=" --weight-decay 0.01"
|
|
OPTS+=" --clip-grad 1.0"
|
|
OPTS+=" --loss-scale 1048576"
|
|
OPTS+=" --max-loss-scale 33554432"
|
|
OPTS+=" --min-loss-scale 1"
|
|
OPTS+=" --loss-scale-steps 32"
|
|
|
|
OPTS+=" --offload"
|
|
OPTS+=" --batch-size 2"
|
|
OPTS+=" --max-length 4096"
|
|
OPTS+=" --lr 3e-4"
|
|
OPTS+=" --start-step 0"
|
|
OPTS+=" --epoch 4"
|
|
OPTS+=" --load /v2/sft_8b_v2/cpm_live_8b-1500-float16.pt"
|
|
OPTS+=" --dataset /workspace/repo/CPM-9G-8B/dataset_bin"
|
|
# TODO 这些 /data 在启元机器上需要改成 /home 下的路径
|
|
OPTS+=" --save ${EXP_PATH}/checkpoints"
|
|
OPTS+=" --save-name ${MODEL_NAME}"
|
|
OPTS+=" --tensorboard ${TB_PATH}"
|
|
|
|
OPTS+=" --delta-tuning"
|
|
OPTS+=" --delta-type lora"
|
|
OPTS+=" --lora-r 64" # 常用的lora 参数
|
|
OPTS+=" --lora-dropout 0.05"
|
|
OPTS+=" --lora-alpha 64" # 常用的lora alpha 参数
|
|
OPTS+=" --lora-layer project_q project_v project_k w_0 w_1 w_out"
|
|
OPTS+=" --save-origin-model"
|
|
|
|
OPTS+=" $@"
|
|
|
|
CMD="torchrun --nnodes=1 --nproc_per_node=6 --rdzv_id=1 --rdzv_backend=c10d --rdzv_endpoint=${MASTER_ADDR}:${MASTER_PORT} ${CPM_PATH}/apps/cpm9g/sft_cpm9g_delta.py ${OPTS}"
|
|
|
|
echo "${CMD}"
|
|
$CMD |