meta-llama

/

Llama-4-Scout-17B-16E-Instruct

like 1.02k

Follow

Meta Llama 54.3k

https://github.com/modelscope/ms-swift/pull/3777

SFT shell:

CUDA_VISIBLE_DEVICES=0,1,2,3 \
swift sft \
    --model LLM-Research/Llama-4-Scout-17B-16E-Instruct \
    --dataset 'AI-ModelScope/LaTeX_OCR:human_handwrite#2000' \
    --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_modules all-linear \
    --freeze_vit true \
    --gradient_accumulation_steps 16 \
    --gradient_checkpointing false \
    --eval_steps 50 \
    --save_steps 50 \
    --save_total_limit 2 \
    --logging_steps 5 \
    --max_length 2048 \
    --output_dir output \
    --warmup_ratio 0.05 \
    --dataloader_num_workers 4