ValueError: Adapter pissa_init does not exist

[xinghaow99]

Reminder

• I have read the README and searched the existing issues.

System Info

• llamafactory version: 0.8.4.dev0
• Platform: Linux-5.4.119-19.0009.28-x86_64-with-glibc2.31
• Python version: 3.10.14
• PyTorch version: 2.4.0+cu121 (GPU)
• Transformers version: 4.43.3
• Datasets version: 2.20.0
• Accelerate version: 0.33.0
• PEFT version: 0.12.0
• TRL version: 0.9.6
• GPU type: NVIDIA H800
• DeepSpeed version: 0.14.4

Reproduction

### model
model_name_or_path: meta-llama/Llama-2-7b-hf

### method
stage: sft
do_train: true
finetuning_type: lora
lora_target: all
pissa_init: true
pissa_iter: 16
pissa_convert: true

### dataset
dataset_dir: LLaMA-Factory/data
dataset: identity,alpaca_en_demo
template: llama2
cutoff_len: 1024
max_samples: 100
overwrite_cache: true
preprocessing_num_workers: 16

### output
output_dir: ckpts/pissa_sft/llama-2-7b
logging_steps: 10
save_steps: 500
plot_loss: true
overwrite_output_dir: true

### train
per_device_train_batch_size: 1
gradient_accumulation_steps: 8
learning_rate: 1.0e-4
num_train_epochs: 3.0
lr_scheduler_type: cosine
warmup_ratio: 0.1
bf16: true
ddp_timeout: 180000000

examples中pissa样例，报错：

Traceback (most recent call last):
  File "/home/jovyan/wxh/LLaMA-Factory/src/llamafactory/launcher.py", line 23, in <module>
    launch()
  File "/home/jovyan/wxh/LLaMA-Factory/src/llamafactory/launcher.py", line 19, in launch
    run_exp()
  File "/home/jovyan/wxh/LLaMA-Factory/src/llamafactory/train/tuner.py", line 50, in run_exp
    run_sft(model_args, data_args, training_args, finetuning_args, generating_args, callbacks)
  File "/home/jovyan/wxh/LLaMA-Factory/src/llamafactory/train/sft/workflow.py", line 94, in run_sft
    train_result = trainer.train(resume_from_checkpoint=training_args.resume_from_checkpoint)
  File "/home/jovyan/conda-env/envs/wxh_lf/lib/python3.10/site-packages/transformers/trainer.py", line 1938, in train
    return inner_training_loop(
  File "/home/jovyan/conda-env/envs/wxh_lf/lib/python3.10/site-packages/transformers/trainer.py", line 2438, in _inner_training_loop
    self.control = self.callback_handler.on_train_end(args, self.state, self.control)
  File "/home/jovyan/conda-env/envs/wxh_lf/lib/python3.10/site-packages/transformers/trainer_callback.py", line 463, in on_train_end
    return self.call_event("on_train_end", args, state, control)
  File "/home/jovyan/conda-env/envs/wxh_lf/lib/python3.10/site-packages/transformers/trainer_callback.py", line 507, in call_event
    result = getattr(callback, event)(
  File "/home/jovyan/wxh/LLaMA-Factory/src/llamafactory/train/callbacks.py", line 168, in on_train_end
    model.delete_adapter("pissa_init")
  File "/home/jovyan/conda-env/envs/wxh_lf/lib/python3.10/site-packages/peft/tuners/lora/model.py", line 821, in delete_adapter
    raise ValueError(f"Adapter {adapter_name} does not exist")
ValueError: Adapter pissa_init does not exist

Expected behavior

No response

Others

No response

[codemayq]

临时你可以 按照 这个 https://github.com/hiyouga/LLaMA-Factory/pull/4995， 手动修复一下。
另外 在推理的时候需要指定 lora 位置如下， 否则可能会出错

adapter_name_or_path: saves/llama3-8b/lora/sft/pissa_converted

[keakon-pureglobal]

@codemayq @hiyouga 修改后在推理或导出时会报 size mismatch：

  File "/home/ps/.pyenv/versions/3.10.14/envs/train/lib/python3.10/site-packages/peft/utils/save_and_load.py", line 395, in set_peft_model_state_dict
    load_result = model.load_state_dict(peft_model_state_dict, strict=False)
  File "/home/ps/.pyenv/versions/3.10.14/envs/train/lib/python3.10/site-packages/torch/nn/modules/module.py", line 2215, in load_state_dict
    raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format(
RuntimeError: Error(s) in loading state_dict for PeftModelForCausalLM:
        size mismatch for base_model.model.model.layers.0.attention.wqkv.lora_A.default.weight: copying a param with shape torch.Size([8, 4096]) from checkpoint, the shape in current model is torch.Size([16, 4096]).

用的是 webui，只要用 pissa 微调就出错。

[keakon-pureglobal]

搞定了，自己构造一个 yaml，把 adapter_name_or_path 改成 pissa_converted 的地址，然后执行 llamafactory-cli webchat xxx.yaml