#!/bin/bash

export CUDA_DEVICE_MAX_CONNECTIONS=1


NPUS_PER_NODE=8
MASTER_ADDR=localhost
MASTER_PORT=6024
NNODES=1
NODE_RANK=0
WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES))

basepath=$(cd `dirname $0`; cd ../../../../; pwd)

DATA_PATH="/data/ci/datasets/origin/train-00000-of-00001-a09b74b3ef9c3b56.parquet"
TOKENIZER_MODEL="/data/ci/models/llama3/hf/llama-3-8b-hf/"
CKPT_LOAD_DIR="/data/ci/models/llama3/hf/llama-3-8b-hf/"
MG_CACHE_DIR="/data/ci/cache/llama-3-8b-tp1pp8"
OUTPUT_PREFIX="/data/ci/cache/llama3-8b-dataset/llama3_8b"

TP=1
PP=8

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

GPT_ARGS="
    --stage sft \
    --tensor-model-parallel-size ${TP} \
    --pipeline-model-parallel-size ${PP} \
    --use-mcore-models \
    --micro-batch-size 1 \
    --global-batch-size 8 \
    --sequence-parallel \
    --use-flash-attn \
    --prompt-type llama3 \
    --no-pad-to-seq-lengths \
    --use-rotary-position-embeddings \
    --rope-scaling-type llama3 \
    --rope-scaling-factor 8.0 \
    --rotary-percent 1.0 \
    --low-freq-factor 1.0 \
    --high-freq-factor 4.0 \
    --original-max-position-embeddings 8192 \
    --tokenizer-type PretrainedFromHF \
    --tokenizer-name-or-path ${TOKENIZER_MODEL} \
    --tokenizer-not-use-fast \
    --num-layers 8 \
    --hidden-size 4096 \
    --ffn-hidden-size 14336 \
    --num-attention-heads 32 \
    --group-query-attention \
    --num-query-groups 8 \
    --seq-length 8192 \
    --max-position-embeddings 8192 \
    --make-vocab-size-divisible-by 1 \
    --padded-vocab-size 128256 \
    --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 \
    --rotary-base 500000 \
    --normalization RMSNorm \
    --norm-epsilon 1e-5 \
    --swiglu \
    --no-masked-softmax-fusion \
    --attention-softmax-in-fp32 \
    --lr 1e-6 \
    --train-iters 15 \
    --lr-decay-style cosine \
    --weight-decay 0.0 \
    --clip-grad 1.0 \
    --adam-beta1 0.9 \
    --adam-beta2 0.999 \
    --initial-loss-scale 4096 \
    --finetune \
    --is-instruction-dataset \
    --no-gradient-accumulation-fusion \
    --no-load-optim \
    --no-load-rng \
    --bf16 \
    --seed 42 \
    --vocab-size 128256 \
    --ckpt-format torch
"

FINETUNE_ARGS="
    --lora-r 8 \
    --lora-alpha 16 \
    --lora-fusion \
    --lora-target-modules linear_qkv linear_proj linear_fc1 linear_fc2 \
"

DATA_ARGS="
    --data-path $DATA_PATH \
    --split 100,0,0 \
    --handler-name AlpacaStyleInstructionHandler \
    --output-prefix ${OUTPUT_PREFIX} \
"

OUTPUT_ARGS="
    --log-interval 1 \
    --save-interval 2000 \
    --eval-interval 2000 \
    --eval-iters 10 \
    --load ${CKPT_LOAD_DIR} \
"

torchrun $DISTRIBUTED_ARGS $basepath/posttrain_gpt.py \
    $GPT_ARGS \
    $DATA_ARGS \
    $FINETUNE_ARGS \
    $OUTPUT_ARGS \
    --distributed-backend nccl \
    --enable-hf2mg-convert \
    --model-type-hf llama2 \
    --transformer-impl local \
    --mg-save-dir ${MG_CACHE_DIR} \