finetune_ds_debug.sh

#!/bin/bash
export CUDA_DEVICE_MAX_CONNECTIONS=1
DIR=`pwd`

GPUS_PER_NODE=8
NNODES=1
NODE_RANK=0
MASTER_ADDR=localhost
MASTER_PORT=6001

MODEL="Qwen/Qwen-VL" # We use the first version of Qwen-VL
# ATTENTION: specify the path to your training data, which should be a json file consisting of a list of conversations.
# See the section for finetuning in README for more information.
DATA="pathto/data"

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


torchrun $DISTRIBUTED_ARGS finetune_multitask.py \
    --model_name_or_path $MODEL \
    --data_path $DATA \
    --bf16 True \
    --fix_vit True \
    --output_dir output_model \
    --num_train_epochs 1 \
    --per_device_train_batch_size 1 \
    --per_device_eval_batch_size 1 \
    --gradient_accumulation_steps 16 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 1000 \
    --save_total_limit 10 \
    --learning_rate 1e-5 \
    --weight_decay 0.1 \
    --adam_beta2 0.95 \
    --warmup_ratio 0.02 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --report_to "none" \
    --model_max_length 2048 \
    --gradient_checkpointing \
    --lazy_preprocess True \
    --deepspeed finetune/ds_config_zero2.json