wehub-resource-sync
50 lines
1.4 KiB
Bash
50 lines
1.4 KiB
Bash
# 4 * 30GiB
|
|
PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
|
|
NPROC_PER_NODE=4 \
|
|
IMAGE_MAX_TOKEN_NUM=1024 \
|
|
VIDEO_MAX_TOKEN_NUM=128 \
|
|
FPS_MAX_FRAMES=12 \
|
|
CUDA_VISIBLE_DEVICES=0,1,2,3 \
|
|
swift sft \
|
|
--model Qwen/Qwen3.5-35B-A3B \
|
|
--tuner_type lora \
|
|
--dataset 'AI-ModelScope/LaTeX_OCR:human_handwrite#2000' \
|
|
--load_from_cache_file true \
|
|
--add_non_thinking_prefix true \
|
|
--split_dataset_ratio 0.01 \
|
|
--torch_dtype bfloat16 \
|
|
--num_train_epochs 1 \
|
|
--per_device_train_batch_size 4 \
|
|
--per_device_eval_batch_size 4 \
|
|
--learning_rate 1e-4 \
|
|
--lora_rank 8 \
|
|
--lora_alpha 32 \
|
|
--target_modules all-linear \
|
|
--experts_impl grouped_mm \
|
|
--router_aux_loss_coef 1e-3 \
|
|
--gradient_accumulation_steps 1 \
|
|
--group_by_length true \
|
|
--output_dir output/Qwen3.5-35B-A3B \
|
|
--eval_steps 50 \
|
|
--save_steps 50 \
|
|
--save_total_limit 2 \
|
|
--logging_steps 5 \
|
|
--max_length 2048 \
|
|
--warmup_ratio 0.05 \
|
|
--dataset_num_proc 4 \
|
|
--dataloader_num_workers 4 \
|
|
--deepspeed zero3
|
|
|
|
|
|
# PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True' \
|
|
# CUDA_VISIBLE_DEVICES=0,1,2,3 \
|
|
# IMAGE_MAX_TOKEN_NUM=1024 \
|
|
# VIDEO_MAX_TOKEN_NUM=128 \
|
|
# FPS_MAX_FRAMES=12 \
|
|
# swift infer \
|
|
# --adapters output/Qwen3.5-35B-A3B/vx-xxx/checkpoint-xxx \
|
|
# --stream true \
|
|
# --experts_impl grouped_mm \
|
|
# --enable_thinking false \
|
|
# --load_data_args true
|