pretrain_t5_distributed_with_mp.sh

#!/bin/bash

GPUS_PER_NODE=8
# Change for multinode config
MASTER_ADDR=localhost
MASTER_PORT=6000
NNODES=1
NODE_RANK=0
WORLD_SIZE=$(($GPUS_PER_NODE*$NNODES))

DATA_PATH=<Specify path and file prefix>
CHECKPOINT_PATH=<Specify path>

DISTRIBUTED_ARGS="--nproc_per_node $GPUS_PER_NODE --nnodes $NNODES --node_rank $NODE_RANK --master_addr $MASTER_ADDR --master_port $MASTER_PORT"

python -m torch.distributed.launch $DISTRIBUTED_ARGS \
       pretrain_t5.py \
       --tensor-model-parallel-size 2 \
       --num-layers 12 \
       --hidden-size 768 \
       --num-attention-heads 12 \
       --kv-channels 64 \
       --ffn-hidden-size 3072 \
       --encoder-seq-length 512 \
       --decoder-seq-length 128 \
       --micro-batch-size 16 \
       --global-batch-size 2048 \
       --seq-length 512 \
       --max-position-embeddings 512 \
       --train-iters 1000000 \
       --lr-decay-iters 1000000 \
       --save $CHECKPOINT_PATH \
       --load $CHECKPOINT_PATH \
       --data-path $DATA_PATH \
       --vocab-file t5-vocab.txt \
       --data-impl mmap \
       --split 949,50,1 \
       --lr 0.0001 \
       --min-lr 0.00001 \
       --lr-decay-style linear \
       --lr-warmup-fraction .01 \
       --weight-decay 1e-2 \
       --clip-grad 1.0 \
       --log-interval 100 \
       --save-interval 10000 \
       --eval-interval 1000 \
       --eval-iters 10 \
       --fp16