wehub-resource-sync
32 lines
898 B
Bash
32 lines
898 B
Bash
CUDA_VISIBLE_DEVICES=0 \
|
|
IMAGE_MAX_TOKEN_NUM=1024 \
|
|
VIDEO_MAX_TOKEN_NUM=128 \
|
|
FPS_MAX_FRAMES=16 \
|
|
swift sft \
|
|
--model Qwen/Qwen3-VL-4B-Instruct \
|
|
--tuner_type lora \
|
|
--dataset 'clip-benchmark/wds_voc2007_multilabel' \
|
|
--load_from_cache_file true \
|
|
--split_dataset_ratio 0.01 \
|
|
--torch_dtype bfloat16 \
|
|
--num_train_epochs 2 \
|
|
--per_device_train_batch_size 16 \
|
|
--per_device_eval_batch_size 16 \
|
|
--learning_rate 1e-4 \
|
|
--lora_rank 8 \
|
|
--lora_alpha 32 \
|
|
--target_modules all-linear \
|
|
--gradient_accumulation_steps 1 \
|
|
--eval_steps 100 \
|
|
--save_steps 100 \
|
|
--save_total_limit 2 \
|
|
--logging_steps 5 \
|
|
--max_length 2048 \
|
|
--output_dir output \
|
|
--warmup_ratio 0.05 \
|
|
--dataloader_num_workers 4 \
|
|
--dataset_num_proc 4 \
|
|
--num_labels 20 \
|
|
--task_type seq_cls \
|
|
--problem_type multi_label_classification
|