#!/bin/bash
export CUDA_DEVICE_MAX_CONNECTIONS=1
export CPU_AFFINITY_CONF=1
export TASK_QUEUE_ENABLE=2
export PYTORCH_NPU_ALLOC_CONF="expandable_segments:True"
export HCCL_CONNECT_TIMEOUT=3600
export STREAMS_PER_DEVICE=32
NPUS_PER_NODE=8
MASTER_ADDR=localhost
MASTER_PORT=6000
NNODES=1
NODE_RANK=0
WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES))
basepath=$(cd `dirname $0`; cd ../../../; pwd)
DATA_PATH=/data/ci/datasets/processed/pretrain_dataset/alpaca_text_document
TOKENIZER_PATH=/data/ci/models/deepseek3/mg/deepseek-v3-mcore-tp1-pp2-ep4-16experts
CKPT_LOAD_DIR=/data/ci/models/deepseek32/mg/deepseek32-tp1-pp2-ep4-8expert
TP=1
PP=2
EP=4
CP=1
CP_TYPE='ulysses_cp_algo'
NUM_LAYERS=4
SEQ_LEN=2048
MBS=1
GBS=16
DISTRIBUTED_ARGS="
--nproc_per_node $NPUS_PER_NODE \
--nnodes $NNODES \
--node_rank $NODE_RANK \
--master_addr $MASTER_ADDR \
--master_port $MASTER_PORT
"
MLA_ARGS="
--multi-latent-attention \
--qk-pos-emb-head-dim 64 \
--qk-head-dim 128 \
--q-lora-rank 1536 \
--kv-lora-rank 512 \
--v-head-dim 128 \
--qk-layernorm \
--mla-mm-split \
--mla-fa-without-pad \
"
MOE_ARGS="
--moe-grouped-gemm \
--moe-permutation-async-comm \
--moe-token-dispatcher-type alltoall \
--moe-permute-fusion \
--first-k-dense-replace 0 \
--moe-layer-freq 1 \
--moe-shared-expert-intermediate-size 2048 \
--num-experts 8 \
--moe-router-topk 2 \
--moe-ffn-hidden-size 2048 \
--moe-router-load-balancing-type seq_aux_loss \
--moe-router-num-groups 4 \
--moe-router-group-topk 1 \
--moe-router-topk-scaling-factor 2.5 \
--moe-aux-loss-coeff 0.0001 \
--moe-router-score-function sigmoid \
--moe-router-enable-expert-bias \
--moe-router-dtype fp32 \
"
MTP_ARGS="
--mtp-num-layers 1 \
--mtp-loss-scaling-factor 0.3 \
"
PIPELINE_ARGS="
--moe-fb-overlap \
--num-layers-per-virtual-pipeline-stage 1 \
"
MEM_ARGS="
--mtp-mem-efficient-logits \
--swap-optimizer \
--recompute-granularity full \
--recompute-method uniform \
--recompute-num-layers 1 \
"
DSA_ARGS="
--enable-dsa-indexer \
--index-topk 1024 \
--indexer-loss-coeff 1.0 \
--init-norm-weight-in-fp32 \
"
OTHERS_ARGS="
--tensorboard-dir ./tb \
--no-shared-storage \
"
ROPE_ARGS="
--beta-fast 32 \
--beta-slow 1 \
--rope-scaling-factor 40 \
--rope-scaling-mscale 1.0 \
--rope-scaling-mscale-all-dim 1.0 \
--rope-scaling-original-max-position-embeddings 4096 \
--rope-scaling-type yarn
"
GPT_ARGS="
--finetune \
--transformer-impl local \
--spec mindspeed_llm.tasks.models.spec.deepseek_spec layer_spec \
--gemm-gradient-accumulation-fusion \
--noop-layers 3 \
--manual-gc \
--manual-gc-interval 50 \
--use-distributed-optimizer \
--use-flash-attn \
--use-mcore-models \
--tensor-model-parallel-size ${TP} \
--pipeline-model-parallel-size ${PP} \
--expert-model-parallel-size ${EP} \
--expert-tensor-parallel-size 1 \
--sequence-parallel \
--context-parallel-size ${CP} \
--context-parallel-algo ${CP_TYPE} \
--num-layers ${NUM_LAYERS} \
--hidden-size 7168 \
--ffn-hidden-size 18432 \
--num-attention-heads 128 \
--tokenizer-type PretrainedFromHF \
--tokenizer-name-or-path ${TOKENIZER_PATH} \
--seq-length ${SEQ_LEN} \
--max-position-embeddings 163840 \
--micro-batch-size ${MBS} \
--global-batch-size ${GBS} \
--make-vocab-size-divisible-by 1 \
--lr 1.0e-5 \
--train-iters 15 \
--lr-decay-style cosine \
--untie-embeddings-and-output-weights \
--disable-bias-linear \
--attention-dropout 0.0 \
--init-method-std 0.02 \
--hidden-dropout 0.0 \
--position-embedding-type rope \
--normalization RMSNorm \
--use-fused-rotary-pos-emb \
--use-rotary-position-embeddings \
--use-fused-swiglu \
--use-fused-rmsnorm \
--swiglu \
--no-masked-softmax-fusion \
--attention-softmax-in-fp32 \
--min-lr 1.0e-7 \
--weight-decay 1e-2 \
--lr-warmup-iters 0 \
--clip-grad 1.0 \
--adam-beta1 0.9 \
--adam-beta2 0.999 \
--initial-loss-scale 65536 \
--vocab-size 129280 \
--padded-vocab-size 129280 \
--rotary-base 10000 \
--norm-epsilon 1e-6 \
--no-load-optim \
--no-load-rng \
--bf16 \
--distributed-timeout-minutes 120 \
"
DATA_ARGS="
--data-path $DATA_PATH \
--split 100,0,0
"
OUTPUT_ARGS="
--log-throughput \
--log-interval 1 \
--save-interval 2000 \
--eval-interval 2000 \
--eval-iters 0 \
--no-save-optim \
--no-save-rng
"
python -m torch.distributed.launch $DISTRIBUTED_ARGS $basepath/pretrain_gpt.py \
$GPT_ARGS \
$DATA_ARGS \
$OUTPUT_ARGS \
$MLA_ARGS \
$PIPELINE_ARGS \
$MEM_ARGS \
$ROPE_ARGS \
$MOE_ARGS \
$DSA_ARGS \
$OTHERS_ARGS \
--load $CKPT_LOAD_DIR \
--transformer-impl local \
--distributed-backend nccl