ymcui / Chinese-LLaMA-Alpaca-2

中文LLaMA-2 & Alpaca-2大模型二期项目 + 64K超长上下文模型 (Chinese LLaMA-2 & Alpaca-2 LLMs with 64K long context models)
Apache License 2.0
7.11k stars 574 forks source link

微调后的lora模块 #558

Closed ymourenya closed 7 months ago

ymourenya commented 7 months ago

提交前必须检查以下项目

问题类型

模型训练与精调

基础模型

Chinese-LLaMA-2-16K (7B/13B)

操作系统

None

详细描述问题

# 请在此处粘贴运行代码(请粘贴在本代码块里)

依赖情况(代码类问题务必提供)

# 请在此处粘贴依赖情况(请粘贴在本代码块里)

运行日志或截图

你好,为预训练出来的lora为什么才48B,是没保存吗,用的你们预训练的代码,这大小不对把 图片

ymourenya commented 7 months ago

lr=2e-4 lora_rank=8 lora_alpha=32 lora_trainable="q_proj,v_proj,k_proj,o_proj,gate_proj,down_proj,up_proj" modules_to_save="embed_tokens,lm_head" lora_dropout=0.05

pretrained_model=./llama2 chinese_tokenizer_path=./llama2 dataset_dir=./data data_cache=1 per_device_train_batch_size=16 gradient_accumulation_steps=8 block_size=512 output_dir=output_dir

deepspeed_config_file=ds_zero2_no_offload.json

torchrun --nnodes 1 --nproc_per_node 2 run_clm_pt_with_peft.py \ --deepspeed ${deepspeed_config_file} \ --model_name_or_path ${pretrained_model} \ --tokenizer_name_or_path ${chinese_tokenizer_path} \ --dataset_dir ${dataset_dir} \ --data_cache_dir ${data_cache} \ --validation_split_percentage 0.001 \ --per_device_train_batch_size ${per_device_train_batch_size} \ --do_train \ --seed $RANDOM \ --fp16 \ --num_train_epochs 1 \ --lr_scheduler_type cosine \ --learning_rate ${lr} \ --warmup_ratio 0.05 \ --weight_decay 0.01 \ --logging_strategy steps \ --logging_steps 10 \ --save_strategy epoch \ --save_total_limit 1 \ --gradient_accumulation_steps ${gradient_accumulation_steps} \ --preprocessing_num_workers 8 \ --block_size ${block_size} \ --output_dir ${output_dir} \ --overwrite_output_dir \ --ddp_timeout 30000 \ --logging_first_step True \ --lora_rank ${lora_rank} \ --lora_alpha ${lora_alpha} \ --trainable ${lora_trainable} \ --lora_dropout ${lora_dropout} \ --modules_to_save ${modules_to_save} \ --torch_dtype float16 \ --load_in_kbits 16 \ --save_safetensors True \ --gradient_checkpointing \ --ddp_find_unused_parameters False

ZEMINGMA commented 7 months ago

已收到,感谢来信。祝好!

ymourenya commented 7 months ago

你好,代码我没有改动任何地方,是哪里有什么问题吗,为什么预训练出来的lora模块才48B

ymourenya commented 7 months ago

已收到,感谢来信。祝好!

你好,能问一下这是什么问题吗

iMountTai commented 7 months ago

pt_lora_model下权重正常吗?

ymourenya commented 7 months ago

pt_lora_model下权重正常吗?

也是这样

Shajiu commented 7 months ago

你好,代码我没有改动任何地方,是哪里有什么问题吗,为什么预训练出来的lora模块才48B

兄弟,这个坑在这里,你只需要在训练脚本中注释如下代码即可正常:

# old_state_dict = model.state_dict
        # model.state_dict = (
        #     lambda self, *_, **__: get_peft_model_state_dict(self, old_state_dict())
        # ).__get__(model, type(model))
github-actions[bot] commented 7 months ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your consideration.

github-actions[bot] commented 7 months ago

Closing the issue, since no updates observed. Feel free to re-open if you need any further assistance.