pretrain_plm_1point8b_ptd.sh

#!/bin/bash
export HCCL_CONNECT_TIMEOUT=1800
export CUDA_DEVICE_MAX_CONNECTIONS=1
export NPU_ASD_ENABLE=0
export HCCL_BUFFSIZE=64
export CPU_AFFINITY_CONF=1,lazy_bind:0

NPUS_PER_NODE=8
MASTER_ADDR=localhost
MASTER_PORT=6006
NNODES=1
NODE_RANK=0
WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES))

# please fill these path configurations
CKPT_SAVE_DIR="your model save ckpt path"
DATA_PATH="your data path"
TOKENIZER_MODEL="your tokenizer path"
CKPT_LOAD_DIR="your model ckpt path"

TP=1
PP=1

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

MLA_ARGS="
    --spec mindspeed_llm.tasks.models.spec.plm_spec layer_spec \
    --multi-latent-attention \
    --qk-pos-emb-head-dim 64 \
    --qk-head-dim 128 \
    --kv-lora-rank 512 \
    --v-head-dim 128 \
"

ROPE_ARGS="
    --rope-scaling-type plm
"

GPT_ARGS="
    --load $CKPT_LOAD_DIR \
    --use-distributed-optimizer \
    --overlap-grad-reduce \
    --overlap-param-gather \
    --use-flash-attn \
    --use-mcore-models \
    --tensor-model-parallel-size ${TP} \
    --pipeline-model-parallel-size ${PP} \
    --sequence-parallel \
    --num-layers 32 \
    --hidden-size 2048 \
    --ffn-hidden-size 8192 \
    --num-attention-heads 16 \
    --num-query-groups 16 \
    --tokenizer-type PretrainedFromHF  \
    --tokenizer-name-or-path ${TOKENIZER_MODEL} \
    --seq-length 4096 \
    --max-position-embeddings 4096 \
    --micro-batch-size 2 \
    --global-batch-size 16 \
    --make-vocab-size-divisible-by 1 \
    --lr 1.25e-06 \
    --train-iters 2000 \
    --lr-decay-style constant \
    --lr-decay-iters 2000 \
    --disable-bias-linear \
    --attention-dropout 0 \
    --init-method-std 0.008 \
    --hidden-dropout 0 \
    --position-embedding-type rope \
    --rotary-base 100000 \
    --normalization RMSNorm \
    --use-fused-rotary-pos-emb \
    --use-rotary-position-embeddings \
    --use-fused-rmsnorm \
    --squared-relu \
    --no-masked-softmax-fusion \
    --attention-softmax-in-fp32 \
    --weight-decay 0.1 \
    --clip-grad 1.0 \
    --adam-beta1 0.9 \
    --adam-beta2 0.999 \
    --initial-loss-scale 1 \
    --vocab-size 151936 \
    --padded-vocab-size 151936 \
    --no-gradient-accumulation-fusion \
    --norm-epsilon 1e-6 \
    --no-load-optim \
    --no-load-rng \
    --bf16
"

DATA_ARGS="
    --data-path $DATA_PATH \
    --split 100,0,0
"

OUTPUT_ARGS="
    --log-interval 1 \
    --save-interval 2000 \
    --eval-interval 1000 \
    --eval-iters 0 \
    --no-save-optim \
    --no-save-rng
"

python -m torch.distributed.launch $DISTRIBUTED_ARGS pretrain_gpt.py \
    $GPT_ARGS \
    $DATA_ARGS \
    $OUTPUT_ARGS \
    $MLA_ARGS \
    $ROPE_ARGS \
    --distributed-backend nccl \
    --save $CKPT_SAVE_DIR \
    --transformer-impl local \
    | tee ./logs/train_plm_1point8b.log