File size: 2,273 Bytes
dc71c15 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 |
#!/bin/bash
# Environment Variables
ARG_WORLD_SIZE=${1:-1}
ARG_NPROC_PER_NODE=${2:-8}
ARG_MASTER_ADDR="127.0.0.1"
ARG_MASTER_PORT=16666
ARG_RANK=0
# Multiple conditions
if [ ! -n "$WORLD_SIZE" ] || [ ! -n "$NPROC_PER_NODE" ]; then
WORLD_SIZE=$ARG_WORLD_SIZE
NPROC_PER_NODE=$ARG_NPROC_PER_NODE
fi
if [ ! -n "$MASTER_ADDR" ] || [ ! -n "$MASTER_PORT" ] || [ ! -n "$RANK" ]; then
MASTER_ADDR=$ARG_MASTER_ADDR
MASTER_PORT=$ARG_MASTER_PORT
RANK=$ARG_RANK
fi
echo "WORLD_SIZE: $WORLD_SIZE"
echo "NPROC_PER_NODE: $NPROC_PER_NODE"
# Training Arguments
GLOBAL_BATCH_SIZE=256
LOCAL_BATCH_SIZE=8
GRADIENT_ACCUMULATION_STEPS=$[$GLOBAL_BATCH_SIZE/($WORLD_SIZE*$NPROC_PER_NODE*$LOCAL_BATCH_SIZE)]
echo $GRADIENT_ACCUMULATION_STEPS
# Log Arguments
export TRANSFORMERS_OFFLINE=1
export WANDB_PROJECT=videollama2mistral_siglip
RUN_NAME=vllava_settings
DATA_DIR=datasets
OUTP_DIR=work_dirs
torchrun --nnodes $WORLD_SIZE \
--nproc_per_node $NPROC_PER_NODE \
--master_addr=$MASTER_ADDR \
--master_port=$MASTER_PORT \
--node_rank $RANK \
videollama2/train_flash_attn.py \
--deepspeed scripts/zero3.json \
--model_type videollama2 \
--model_path mistralai/Mistral-7B-Instruct-v0.2 \
--vision_tower google/siglip-so400m-patch14-384 \
--mm_projector_type stc_connector_v35 \
--tune_mm_mlp_adapter True \
--data_path ${DATA_DIR}/videollava_pt/valley_llavaimage.json \
--data_folder ${DATA_DIR}/videollava_pt/ \
--mm_vision_select_layer -2 \
--num_frames 8 \
--bf16 True \
--tf32 True \
--fp16 False \
--output_dir ${OUTP_DIR}/${WANDB_PROJECT}/pretrain_${RUN_NAME} \
--num_train_epochs 1 \
--per_device_train_batch_size $LOCAL_BATCH_SIZE \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps $GRADIENT_ACCUMULATION_STEPS \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 500 \
--save_total_limit 99 \
--learning_rate 1e-3 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--model_max_length 2048 \
--gradient_checkpointing True \
--dataloader_num_workers 16 \
--lazy_preprocess True \
--report_to tensorboard \
--run_name pretrain_$RUN_NAME \
|