File size: 1,645 Bytes
bbfa6f6 |
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 |
#!/bin/bash
deepspeed --num_nodes=1 --num_gpus=7 --master_port=25001 llava/train/train_mem.py \
--deepspeed ./scripts/zero2.json \
--model_name_or_path mistralai/Mistral-7B-Instruct-v0.1 \
--version v1 \
--dataset_config /mnt/bn/algo-masp-nas-2/xiangchen/repo/LLaVA/llava/configs/adso_increasing_ablation/finetune_gpt4v_adso185k.yaml \
--vision_tower eva-vit-g \
--vit_model_path /mnt/bn/data-tns-algo-masp/baiyi.by/masp/model/eva_vit_g.pth \
--qformer_model_path /mnt/bn/data-tns-algo-masp/baiyi.by/masp/model/blip2_pretrained_flant5xxl.pth \
--pretrain_mm_mlp_adapter /mnt/bn/algo-masp-nas-2/xiangchen/model/masp_models/checkpoints/mistral-videoproj-pretrain-frames-base-intvid/mm_projector.bin \
--freeze_qformer False \
--mm_vision_select_layer -2 \
--mm_use_start_end True \
--mm_use_patch_token False \
--image_aspect_ratio pad \
--bf16 True \
--output_dir /mnt/bn/algo-masp-nas-2/xiangchen/model/masp_models/checkpoints/llava-mistral_gpt4v_adso185k_unfreeze_qformer_data_sampler \
--group_by_modality_length True \
--num_train_epochs 1 \
--per_device_train_batch_size 16 \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps 1 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--tf32 True \
--model_max_length 4096 \
--gradient_checkpointing True \
--dataloader_num_workers 1 \
--lazy_preprocess True \
--report_to none
|