finetune_csl_distributed.sh

#!/bin/bash

WORLD_SIZE=8

DISTRIBUTED_ARGS="--nproc_per_node $WORLD_SIZE \
                  --nnodes 1 \
                  --node_rank 0 \
                  --master_addr localhost \
                  --master_port 6000"

TASK="CSL"
TRAIN_DATA="clue_data/csl/train.json"
VALID_DATA="clue_data/csl/dev.json"
TEST_DATA="clue_data/csl/test.json"
PRETRAINED_CHECKPOINT="./yuyan-10b"

VOCAB_FILE=bert-vocab.txt

for lr in 4e-6 7e-6; do
    for bs in 4 2; do
        for ep in 7 10; do
            ct=`date +"%m%d%H%M%S"`
            OUTPUTS_PATH="outputs/${TASK}/yuyan_bs_${bs}_lr_${lr}_ep_${ep}_${ct}"
            if [ ! -d ${OUTPUTS_PATH}  ];then
                mkdir -p ${OUTPUTS_PATH}
            else
                echo "dir exist, not mkdir"
            fi
            python -m torch.distributed.launch $DISTRIBUTED_ARGS ./tasks/main.py \
                        --task $TASK \
                        --seed 1234 \
                        --pretrained-checkpoint $PRETRAINED_CHECKPOINT \
                        --train-data $TRAIN_DATA \
                        --valid-data $VALID_DATA \
                        --test-data $TEST_DATA \
                        --tokenizer-type BertWordPieceLowerCase \
                        --vocab-file $VOCAB_FILE \
                        --epochs $ep \
                        --tensor-model-parallel-size 8 \
                        --num-layers 48 \
                        --hidden-size 4096 \
                        --num-attention-heads 64 \
                        --micro-batch-size $bs \
                        --lr $lr \
                        --lr-decay-style linear \
                        --lr-warmup-fraction 0.1 \
                        --seq-length 512 \
                        --max-position-embeddings 512 \
                        --log-interval 10 \
                        --eval-interval 3000 \
                        --eval-iters 50 \
                        --weight-decay 1.0e-1 \
                        --res-path ${OUTPUTS_PATH} \
                        --fp16 | tee ${OUTPUTS_PATH}/job.log
            #               --activations-checkpoint-method uniform \
                        

        done
    done
done