wehub-resource-sync
33 lines
997 B
Bash
33 lines
997 B
Bash
# If you don't want to train the router, set:
|
|
# `--target_regex '^(language_model).*\.(q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj)$'`
|
|
NPROC_PER_NODE=4 \
|
|
USE_HF=1 \
|
|
CUDA_VISIBLE_DEVICES=0,1,2,3 \
|
|
swift sft \
|
|
--model meta-llama/Llama-4-Scout-17B-16E-Instruct \
|
|
--dataset 'linxy/LaTeX_OCR:full#5000' \
|
|
--load_from_cache_file true \
|
|
--split_dataset_ratio 0.01 \
|
|
--tuner_type lora \
|
|
--torch_dtype bfloat16 \
|
|
--num_train_epochs 1 \
|
|
--per_device_train_batch_size 1 \
|
|
--per_device_eval_batch_size 1 \
|
|
--learning_rate 1e-4 \
|
|
--lora_rank 8 \
|
|
--lora_alpha 32 \
|
|
--router_aux_loss_coef 1e-3 \
|
|
--freeze_vit true \
|
|
--freeze_aligner true \
|
|
--gradient_accumulation_steps 4 \
|
|
--gradient_checkpointing true \
|
|
--eval_steps 50 \
|
|
--save_steps 50 \
|
|
--save_total_limit 2 \
|
|
--logging_steps 5 \
|
|
--max_length 2048 \
|
|
--output_dir output \
|
|
--warmup_ratio 0.05 \
|
|
--deepspeed zero3 \
|
|
--dataloader_num_workers 4
|