#!/bin/bash
export CUDA_DEVICE_MAX_CONNECTIONS=1
NPUS_PER_NODE=8
MASTER_ADDR=localhost
MASTER_PORT=6011
NNODES=1
NODE_RANK=0
WORLD_SIZE=$(($NPUS_PER_NODE * $NNODES))
basepath=$(cd `dirname $0`; cd ../../../; pwd)
DISTRIBUTED_ARGS="
--nproc_per_node $NPUS_PER_NODE \
--nnodes $NNODES \
--node_rank $NODE_RANK \
--master_addr $MASTER_ADDR \
--master_port $MASTER_PORT
"
echo "NODE_RANK ${NODE_RANK}"
DATA_PATH="/data/ci/datasets/processed/qwen3_30b_dist/alpaca_text_document"
TOKENIZER_PATH="/data/ci/models/Qwen3-30B-A3B/hf/Qwen3-30B-A3B"
CKPT_LOAD_DIR="/data/ci/models/Qwen3-30B-A3B/mg/qwen3-30b-layer1-tp4pp1ep2"
TP=4
PP=1
EP=2
CP=2
CP_TYPE='ulysses_cp_algo'
NUM_LAYERS=1
MOE_ARGS="
--num-experts 128 \
--moe-router-topk 8 \
--moe-router-load-balancing-type aux_loss \
--moe-ffn-hidden-size 768 \
--moe-grouped-gemm \
--moe-permutation-async-comm \
--moe-token-dispatcher-type alltoall_seq \
--moe-layer-freq -1 \
--first-k-dense-replace -1 \
--moe-aux-loss-coeff 0.001
"
OPTIMIZE_ARGS="
--sequence-parallel \
--use-distributed-optimizer \
--use-flash-attn \
--use-fused-rotary-pos-emb \
--use-rotary-position-embeddings \
--use-fused-swiglu \
--use-fused-rmsnorm \
--no-masked-softmax-fusion \
--gemm-gradient-accumulation-fusion \
--recompute-method uniform \
--recompute-granularity full \
--recompute-num-layers 1
"
MODEL_PARALLEL_ARGS="
--tensor-model-parallel-size ${TP} \
--pipeline-model-parallel-size ${PP} \
--expert-model-parallel-size ${EP} \
--context-parallel-size ${CP} \
--context-parallel-algo ${CP_TYPE}
"
TRAIN_ARGS="
--micro-batch-size 1 \
--global-batch-size 16 \
--lr 1.25e-6 \
--train-iters 15 \
--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 \
--bf16 \
--seq-length 4096 \
--no-shared-storage \
--manual-gc \
--manual-gc-interval 15
"
GPT_ARGS="
--use-mcore-models \
--spec mindspeed_llm.tasks.models.spec.qwen3_spec layer_spec \
--kv-channels 128 \
--qk-layernorm \
--norm-topk-prob \
--tokenizer-name-or-path ${TOKENIZER_PATH} \
--load ${CKPT_LOAD_DIR}
--max-position-embeddings 32768 \
--num-layers ${NUM_LAYERS} \
--hidden-size 2048 \
--ffn-hidden-size 6144 \
--num-attention-heads 32 \
--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 \
--norm-epsilon 1e-6 \
--swiglu \
--attention-softmax-in-fp32 \
--no-gradient-accumulation-fusion \
--group-query-attention \
--num-query-groups 4
"
DATA_ARGS=(
--data-path $DATA_PATH \
--split 100,0,0
)
OUTPUT_ARGS=(
--finetune \
--log-interval 1 \
--save-interval 10000 \
--eval-interval 10000 \
--eval-iters 0 \
--no-load-optim \
--no-load-rng
)
torchrun ${DISTRIBUTED_ARGS[@]} $basepath/pretrain_gpt.py \
${MOE_ARGS[@]} \
${OPTIMIZE_ARGS[@]}\
${MODEL_PARALLEL_ARGS[@]}\
${TRAIN_ARGS[@]}\
${GPT_ARGS[@]} \
${DATA_ARGS[@]} \
${OUTPUT_ARGS[@]} \
--transformer-impl local \
--ckpt-format torch \
--distributed-backend nccl