feat✨: 更新训练脚本以支持MOELORA,调整梯度累积步数,优化配置文件
This commit is contained in:
Executable
+24
@@ -0,0 +1,24 @@
|
||||
#!/bin/bash
|
||||
|
||||
accelerate launch --config_file configs/accelerate_configs/deepspeed_zero1.yaml train.py \
|
||||
--dataset_name textvqa \
|
||||
--use_peft \
|
||||
--peft_type MOELORA \
|
||||
--model_name_or_path Qwen/Qwen2.5-Omni-3B \
|
||||
--lora_target_modules .*model\.layers.*proj \
|
||||
--lora_r 8 \
|
||||
--lora_alpha 32 \
|
||||
--per_device_train_batch_size 3 \
|
||||
--per_device_eval_batch_size 1 \
|
||||
--gradient_accumulation_steps 2 \
|
||||
--num_train_epochs 1 \
|
||||
--output_dir checkpoint/qwen2_alllinear/ \
|
||||
--learning_rate 2e-4 \
|
||||
--warmup_ratio 0.03 \
|
||||
--lr_scheduler_type cosine \
|
||||
--bf16 \
|
||||
--torch_dtype bfloat16 \
|
||||
--logging_steps 10 \
|
||||
--gradient_checkpointing \
|
||||
--weight_decay 0.1 \
|
||||
--resume_from_checkpoint /root/autodl-tmp/zhouyunyao/projects/CL-LMM/src/checkpoint/qwen2_alllinear/checkpoint-1000
|
||||
Reference in New Issue
Block a user