export HCCL_CONNECT_TIMEOUT=1800
export CUDA_DEVICE_MAX_CONNECTIONS=1

NPUS_PER_NODE=16
MASTER_ADDR=localhost
MASTER_PORT=6000
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_PATH="your tokenizer path"
CKPT_LOAD_DIR="your model ckpt path"

TP=8
PP=2
VPP=4

MBS=2
GBS=256
SEQ_LENGTH=4096
TRAIN_ITERS=2000

DISTRIBUTED_ARGS="
    --local_worker_num $NPUS_PER_NODE \
    --worker_num $WORLD_SIZE \
    --node_rank $NODE_RANK \
    --master_addr $MASTER_ADDR \
    --master_port $MASTER_PORT \
    --join=True
"

OPTIMIZE_ARGS="
    --use-flash-attn \
    --use-fused-rotary-pos-emb \
    --use-rotary-position-embeddings \
    --use-fused-swiglu \
    --use-fused-rmsnorm \
    --no-masked-softmax-fusion \
    --use-distributed-optimizer \
    --overlap-grad-reduce \
    --overlap-param-gather
"

TRAIN_ARGS="
    --micro-batch-size ${MBS} \
    --global-batch-size ${GBS} \
    --lr 1.25e-6 \
    --lr-decay-style cosine \
    --min-lr 1.25e-7 \
    --weight-decay 1e-1 \
    --lr-warmup-fraction 0.01 \
    --attention-dropout 0.0 \
    --init-method-std 0.01 \
    --hidden-dropout 0.0 \
    --clip-grad 1.0 \
    --adam-beta1 0.9 \
    --adam-beta2 0.95 \
    --initial-loss-scale 4096 \
    --seed 42 \
    --bf16 \
    --train-iters ${TRAIN_ITERS} \
    --seq-length ${SEQ_LENGTH} \
    --no-shared-storage
"

MODEL_PARALLEL_ARGS="
    --tensor-model-parallel-size ${TP} \
    --pipeline-model-parallel-size ${PP} \
    --num-layers-per-virtual-pipeline-stage ${VPP} \
"

GPT_ARGS="
    --use-mcore-models \
    --sequence-parallel \
    --spec mindspeed_llm.tasks.models.spec.qwen3_spec layer_spec \
    --kv-channels 128 \
    --qk-layernorm \
    --tokenizer-name-or-path ${TOKENIZER_PATH} \
    --max-position-embeddings ${SEQ_LENGTH} \
    --num-layers 64 \
    --hidden-size 5120 \
    --ffn-hidden-size 25600 \
    --num-attention-heads 64 \
    --tokenizer-type PretrainedFromHF \
    --make-vocab-size-divisible-by 1 \
    --padded-vocab-size 151936 \
    --rotary-base 1000000 \
    --untie-embeddings-and-output-weights \
    --disable-bias-linear \
    --position-embedding-type rope \
    --normalization RMSNorm \
    --swiglu \
    --attention-softmax-in-fp32 \
    --no-gradient-accumulation-fusion \
    --group-query-attention \
    --num-query-groups 8 \
    --reset-attention-mask \
    --use-ascend-coc \
    --coc-fused-kernel \
    --transformer-impl local \
    --ckpt-format torch
"

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

OUTPUT_ARGS="
    --log-interval 1 \
    --save-interval ${TRAIN_ITERS} \
    --eval-interval ${TRAIN_ITERS} \
    --eval-iters 0 \
    --no-load-optim \
    --no-load-rng \
    --log-throughput
"

msrun $DISTRIBUTED_ARGS pretrain_gpt.py \
    $GPT_ARGS \
    $DATA_ARGS \
    $OUTPUT_ARGS \
    $OPTIMIZE_ARGS \
    $TRAIN_ARGS \
    $MODEL_PARALLEL_ARGS \
    --distributed-backend nccl \
    --load ${CKPT_LOAD_DIR} \
    --save ${CKPT_SAVE_DIR} \
    --ai-framework mindspore \
    | tee logs/train_mcore_qwen3_32b_4k_A3.log