File size: 900 Bytes
cb2428f |
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 |
# Manually select `target_modules` to avoid 'all-linear' selecting 'router'
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' \
--train_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 \
--target_regex '^(language_model).*\.(q_proj|k_proj|v_proj|o_proj|gate_proj|up_proj|down_proj)$' \
--freeze_vit 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
|