#!/bin/bash

export CUDA_DEVICE_MAX_CONNECTIONS=1
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True

NPUS_PER_NODE=8
MASTER_ADDR=localhost
MASTER_PORT=6009
NNODES=1
NODE_RANK=0
WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES))

CKPT_SAVE_DIR="your model save ckpt path"
DATA_PATH="your data path"
TOKENIZER_PATH="your tokenizer path"
CKPT_LOAD_DIR="your model ckpt path"
TP=1
PP=1

DISTRIBUTED_ARGS="
    --master_addr $MASTER_ADDR \
    --node_rank $NODE_RANK \
    --worker_num $WORLD_SIZE \
    --local_worker_num $NPUS_PER_NODE \
    --master_port $MASTER_PORT \
    --log_dir=msrun_log \
    --join=False \
    --cluster_time_out=300 \
    --bind_core=True \
"


GPT_ARGS="
    --use-mcore-models \
    --tensor-model-parallel-size ${TP} \
    --pipeline-model-parallel-size ${PP} \
    --sequence-parallel \
    --num-layers 40 \
    --hidden-size 2304 \
    --ffn-hidden-size 5760 \
    --num-attention-heads 36 \
    --tokenizer-type PretrainedFromHF  \
    --tokenizer-name-or-path "${TOKENIZER_PATH}" \
    --tokenizer-not-use-fast \
    --vocab-size 122753 \
    --seq-length 4096 \
    --max-position-embeddings 4096 \
    --micro-batch-size 1 \
    --global-batch-size 8 \
    --make-vocab-size-divisible-by 1 \
    --lr 1e-7 \
    --train-iters 2000 \
    --lr-decay-style constant \
    --disable-bias-linear \
    --attention-dropout 0.0 \
    --init-method-std 0.01 \
    --hidden-dropout 0.0 \
    --position-embedding-type rope \
    --normalization RMSNorm \
    --norm-epsilon 1e-5 \
    --use-fused-rmsnorm \
    --swiglu \
    --use-flash-attn \
    --no-masked-softmax-fusion \
    --attention-softmax-in-fp32 \
    --min-lr 1e-8 \
    --weight-decay 1e-1 \
    --lr-warmup-fraction 0.01 \
    --clip-grad 1.0 \
    --adam-beta1 0.9 \
    --initial-loss-scale 65536 \
    --adam-beta2 0.999 \
    --no-gradient-accumulation-fusion \
    --no-load-optim \
    --no-load-rng \
    --use-distributed-optimizer \
    --use-fused-swiglu \
    --use-fused-rotary-pos-emb \
    --overlap-grad-reduce \
    --variable-seq-lengths \
    --scale-emb 12 \
    --dim-model-base 256 \
    --scale-depth 1.4 \
    --lora-r 8 \
    --lora-alpha 16 \
    --lora-fusion \
    --lora-target-modules linear_qkv linear_proj linear_fc1 linear_fc2 \
    --bf16 \
"

DATA_ARGS="
    --data-path $DATA_PATH \
    --split 100,0,0 \
    --finetune \
    --stage sft \
    --is-instruction-dataset \
    --prompt-type cpm \
    --no-shuffle \
"

OUTPUT_ARGS="
    --log-interval 1 \
    --save-interval 1000 \
    --eval-interval 1000 \
    --eval-iters 10 \
"

msrun $DISTRIBUTED_ARGS posttrain_gpt.py \
    $GPT_ARGS \
    $DATA_ARGS \
    $OUTPUT_ARGS \
    --distributed-backend nccl \
    --ai-framework mindspore \
    --load $CKPT_LOAD_DIR \
    --save $CKPT_SAVE_DIR \
    | tee logs/tune_minicpm_2b_lora_npu.log