ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add multimodal LLM BLIP-2 and InstructBLIP

Browse Source 
Former-commit-id: 67800c565b086f362b8cf131b0c9babaa7a7ebc7
This commit is contained in:
BUAADreamer
2024-04-23 19:22:42 +08:00
parent 0b99b13786
commit 1451297c78
3 changed files with 4 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
examples/mllm/sft_blip2.sh
View File

@@ -14,7 +14,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--overwrite_output_dir \
--cutoff_len 1024 \
--preprocessing_num_workers 16 \
--per_device_train_batch_size 1 \
--per_device_train_batch_size 4 \
--per_device_eval_batch_size 1 \
--gradient_accumulation_steps 8 \
--lr_scheduler_type cosine \
@@ -30,5 +30,4 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--val_size 0.1 \
--plot_loss \
--quantization_bit 8 \
--image_path /home/LAB/fengzc/LLM/checkpoints/liuhaotian/LLaVA-Instruct-150K/images/coco/train2017
--image_path /home/LAB/fengzc/LLM/checkpoints/liuhaotian/LLaVA-Instruct-150K/images/coco/train2017

4
examples/mllm/sft_instructblip.sh
View File

@@ -14,7 +14,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--overwrite_output_dir \
--cutoff_len 1024 \
--preprocessing_num_workers 16 \
--per_device_train_batch_size 1 \
--per_device_train_batch_size 4 \
--per_device_eval_batch_size 1 \
--gradient_accumulation_steps 8 \
--lr_scheduler_type cosine \
@@ -24,7 +24,7 @@ CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
--eval_steps 100 \
--evaluation_strategy steps \
--load_best_model_at_end \
--learning_rate 5e-5 \
--learning_rate 1e-5 \
--num_train_epochs 3.0 \
--max_samples 3000 \
--val_size 0.1 \