#!/bin/bash

# Runs the "7B" parameter model
export HSA_FORCE_FINE_GRAIN_PCIE=1
#export OMP_NUM_THREADS=1
export NCCL_P2P_LEVEL=5

export NCCL_ALGO=Ring
export NCCL_NCHANNELS_PER_PEER=16
export NCCL_MIN_NCHANNELS=20
export CUDA_DEVICE_MAX_CONNECTIONS=1

GPUS_PER_NODE=1 #4
# Change for multinode config
MASTER_ADDR=localhost
MASTER_PORT=6000
NUM_NODES=1
NODE_RANK=0
WORLD_SIZE=$(($GPUS_PER_NODE*$NUM_NODES))

CHECKPOINT_PATH=./tmp #$1 #<Specify path>
TENSORBOARD_LOGS_PATH=./tmp  #$2 #<Specify path>
#VOCAB_FILE=$3 #<Specify path to file>/gpt2-vocab.json
#MERGE_FILE=$4 #<Specify path to file>/gpt2-merges.txt
DATA_PATH="/root/megatron-llama/dataset/my-llama_text_document" #<Specify path and file prefix>_text_document
TOKENIZER_PATH="/root/megatron-llama/tokenizer.model"

DISTRIBUTED_ARGS=(
    --nproc_per_node $GPUS_PER_NODE 
    --nnodes $NUM_NODES 
    --master_addr $MASTER_ADDR 
    --master_port $MASTER_PORT
)

GPT_MODEL_ARGS=(
    --num-layers 16
    --hidden-size 1024 
    --num-attention-heads 16
    --ffn-hidden-size 4096
    --seq-length 2048
    --max-position-embeddings 2048
)

TRAINING_ARGS=(
    --transformer-impl local
    --sequence-parallel
    --micro-batch-size 1 
    --global-batch-size 10  
    --train-iters 5 
    --weight-decay 0.1 
    --adam-beta1 0.9 
    --adam-beta2 0.95 
    --init-method-std 0.006 
    --clip-grad 1.0 
    --bf16
    --use-distributed-optimizer 
    --use-flash-attn-triton 
    --recompute-activations
    --disable-bias-linear
    --attention-dropout 0
    --hidden-dropout 0
    --no-gradient-accumulation-fusion
    --swiglu
    --lr 3.0e-5 
    --lr-decay-style cosine 
    --min-lr 3.0e-6
    --lr-warmup-iters 1
)
#--use-legacy-models
MODEL_PARALLEL_ARGS=(
	--tensor-model-parallel-size 1 
	--pipeline-model-parallel-size 1 
)

DATA_ARGS=(
    --data-path $DATA_PATH 
    --split 949,50,1
    --untie-embeddings-and-output-weights
    --use-rotary-position-embeddings 
    --normalization RMSNorm 
    --no-position-embedding 
    --tokenizer-model $TOKENIZER_PATH 
    --tokenizer-type Llama2Tokenizer
)

EVAL_AND_LOGGING_ARGS=(
    --log-interval 1
    --save-interval 10000 
    --eval-interval 1000 
    --save $CHECKPOINT_PATH 
    --load $CHECKPOINT_PATH 
    --eval-iters 10
    --tensorboard-dir $TENSORBOARD_LOGS_PATH 
)

torchrun ${DISTRIBUTED_ARGS[@]} pretrain_gpt.py \
    ${GPT_MODEL_ARGS[@]} \
    ${TRAINING_ARGS[@]} \
    ${MODEL_PARALLEL_ARGS[@]} \
    ${DATA_ARGS[@]} \
    ${EVAL_AND_LOGGING_ARGS[@]}