Skip to content

Latest commit

 

History

History
94 lines (82 loc) · 2.79 KB

bunny-stablelm-2-eva-lora.md

File metadata and controls

94 lines (82 loc) · 2.79 KB

Training Tutorial of bunny-stablelm-2-eva-lora

Pretrain

#!/bin/bash

MODEL_TYPE=stablelm-2
OUTPUT_DIR=bunny-$MODEL_TYPE-pretrain

mkdir -p ./checkpoints-pretrain/$OUTPUT_DIR

deepspeed bunny/train/train.py \
    --deepspeed ./script/deepspeed/zero2.json \
    --model_name_or_path /path/to/stabilityai/stablelm-2-1_6b \
    --model_type $MODEL_TYPE \
    --version plain \
    --data_path ./data/pretrain/bunny_pretrain_laion_2m.json \
    --image_folder ./data/pretrain/images \
    --vision_tower /path/to/EVA02_CLIP_L_336_psz14_s6B \
    --mm_projector_type mlp2x_gelu \
    --tune_mm_mlp_adapter True \
    --image_aspect_ratio square \
    --bf16 True \
    --output_dir ./checkpoints-pretrain/$OUTPUT_DIR \
    --num_train_epochs 1 \
    --per_device_train_batch_size 8 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 4 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 24000 \
    --save_total_limit 1 \
    --learning_rate 1e-3 \
    --weight_decay 0. \
    --warmup_ratio 0.03 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --tf32 True \
    --model_max_length 2048 \
    --gradient_checkpointing True \
    --dataloader_num_workers 4 \
    --lazy_preprocess True \
    --report_to none | tee 2>&1 ./checkpoints-pretrain/$OUTPUT_DIR/log.txt

Visual Instruction Tuning

#!/bin/bash

MODEL_TYPE=stablelm-2

PRETRAIN_DIR=bunny-$MODEL_TYPE-pretrain
OUTPUT_DIR=bunny-lora-$MODEL_TYPE

mkdir -p ./checkpoints-$MODEL_TYPE/$OUTPUT_DIR

deepspeed bunny/train/train.py \
    --lora_enable True --lora_r 128 --lora_alpha 256 --mm_projector_lr 2e-5 \
    --deepspeed ./script/deepspeed/zero3.json \
    --model_name_or_path /path/to/stabilityai/stablelm-2-1_6b \
    --model_type $MODEL_TYPE \
    --version bunny \
    --data_path ./data/finetune/bunny_695k.json \
    --image_folder ./data/finetune/images \
    --vision_tower /path/to/EVA02_CLIP_L_336_psz14_s6B \
    --pretrain_mm_mlp_adapter ./checkpoints-pretrain/$PRETRAIN_DIR/mm_projector.bin \
    --mm_projector_type mlp2x_gelu \
    --image_aspect_ratio pad \
    --group_by_modality_length False \
    --bf16 True \
    --output_dir ./checkpoints-$MODEL_TYPE/$OUTPUT_DIR \
    --num_train_epochs 1 \
    --per_device_train_batch_size 8 \
    --per_device_eval_batch_size 4 \
    --gradient_accumulation_steps 2 \
    --evaluation_strategy "no" \
    --save_strategy "steps" \
    --save_steps 500 \
    --save_total_limit 1 \
    --learning_rate 2e-4 \
    --weight_decay 0. \
    --warmup_ratio 0.03 \
    --lr_scheduler_type "cosine" \
    --logging_steps 1 \
    --tf32 True \
    --model_max_length 2048 \
    --gradient_checkpointing True \
    --dataloader_num_workers 4 \
    --lazy_preprocess True \
    --report_to none | tee 2>&1 ./checkpoints-$MODEL_TYPE/$OUTPUT_DIR/log.txt