Megatron-DeepSpeed/examples/pretrain_bert_distributed.sh

65 строки
1.3 KiB
Bash
Executable File

#!/bin/bash
export CUDA_DEVICE_MAX_CONNECTIONS=1
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))
CHECKPOINT_PATH=<Specify path>
VOCAB_FILE=<Specify path to file>/bert-vocab.txt
DATA_PATH=<Specify path and file prefix>_text_sentence
DISTRIBUTED_ARGS="
--nproc_per_node $GPUS_PER_NODE \
--nnodes $NNODES \
--node_rank $NODE_RANK \
--master_addr $MASTER_ADDR \
--master_port $MASTER_PORT
"
BERT_ARGS="
--num-layers 24 \
--hidden-size 1024 \
--num-attention-heads 16 \
--seq-length 512 \
--max-position-embeddings 512 \
--micro-batch-size 4 \
--global-batch-size 32 \
--lr 0.0001 \
--train-iters 1000000 \
--lr-decay-iters 990000 \
--lr-decay-style linear \
--min-lr 1.0e-5 \
--weight-decay 1e-2 \
--lr-warmup-fraction .01 \
--clip-grad 1.0 \
--fp16
"
DATA_ARGS="
--data-path $DATA_PATH \
--vocab-file $VOCAB_FILE \
--data-impl mmap \
--split 949,50,1
"
OUTPUT_ARGS="
--log-interval 100 \
--save-interval 10000 \
--eval-interval 1000 \
--eval-iters 10
"
torchrun $DISTRIBUTED_ARGS pretrain_bert.py \
$BERT_ARGS \
$DATA_ARGS \
$OUTPUT_ARGS \
--distributed-backend nccl \
--save $CHECKPOINT_PATH \
--load $CHECKPOINT_PATH