选择pissa微调时，当提示训练完成后，转换为lora时会报错

[therealoliver]

Reminder

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

System Info

Traceback (most recent call last):
File "/usr/local/bin/llamafactory-cli", line 8, in
sys.exit(main())
File "/mnt/workspace/LLaMA-Factory/src/llamafactory/cli.py", line 112, in main
run_exp()
File "/mnt/workspace/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 "/mnt/workspace/LLaMA-Factory/src/llamafactory/train/sft/workflow.py", line 100, in run_sft
train_result = trainer.train(resume_from_checkpoint=training_args.resume_from_checkpoint)
File "/usr/local/lib/python3.10/site-packages/transformers/trainer.py", line 2122, in train
return inner_training_loop(
File "/usr/local/lib/python3.10/site-packages/transformers/trainer.py", line 2628, in _inner_training_loop
self.control = self.callback_handler.on_train_end(args, self.state, self.control)
File "/usr/local/lib/python3.10/site-packages/transformers/trainer_callback.py", line 471, in on_train_end
return self.call_event("on_train_end", args, state, control)
File "/usr/local/lib/python3.10/site-packages/transformers/trainer_callback.py", line 518, in call_event
result = getattr(callback, event)(
File "/mnt/workspace/LLaMA-Factory/src/llamafactory/train/callbacks.py", line 171, in on_train_end
model.save_pretrained(
File "/usr/local/lib/python3.10/site-packages/peft/peft_model.py", line 341, in save_pretrained
output_state_dict = save_mutated_as_lora(
File "/usr/local/lib/python3.10/site-packages/peft/peft_model.py", line 279, in save_mutated_as_lora
self.load_adapter(
File "/usr/local/lib/python3.10/site-packages/peft/peft_model.py", line 1112, in load_adapter
self.add_adapter(adapter_name, peft_config)
File "/usr/local/lib/python3.10/site-packages/peft/peft_model.py", line 873, in add_adapter
self.base_model.inject_adapter(self.base_model.model, adapter_name)
File "/usr/local/lib/python3.10/site-packages/peft/tuners/tuners_utils.py", line 431, in inject_adapter
self._create_and_replace(peft_config, adapter_name, target, target_name, parent, current_key=key)
File "/usr/local/lib/python3.10/site-packages/peft/tuners/lora/model.py", line 224, in _create_and_replace
new_module = self._create_new_module(lora_config, adapter_name, target, **kwargs)
File "/usr/local/lib/python3.10/site-packages/peft/tuners/lora/model.py", line 346, in _create_new_module
raise ValueError(
ValueError: Target module ModuleDict(
(default): Identity()
(pissa_init): Identity()
) is not supported. Currently, only the following modules are supported: torch.nn.Linear, torch.nn.Embedding, torch.nn.Conv2d, transformers.pytorch_utils.Conv1D.

Reproduction

下面是测试的脚本：

bf16: true
cutoff_len: 2048
dataset: mire_train
dataset_dir: data
ddp_timeout: 180000000
do_train: true
finetuning_type: lora
flash_attn: fa2
gradient_accumulation_steps: 2
learning_rate: 5.0e-06
logging_steps: 2
lora_alpha: 16
lora_dropout: 0
lora_rank: 128
lora_target: all
lr_scheduler_type: cosine
max_grad_norm: 1.0
max_samples: 8
model_name_or_path: Qwen/Qwen2-VL-2B-Instruct
model_revision: master
num_train_epochs: 1.0
optim: adamw_torch
output_dir: saves/Qwen2-VL-2B-Instruct/lora/train_short_sample_for_pissa_debug
packing: false
per_device_train_batch_size: 2
pissa_convert: true
pissa_init: true
plot_loss: true
preprocessing_num_workers: 16
report_to: none
save_steps: 2
stage: sft
template: qwen2_vl
warmup_steps: 2

Expected behavior

No response

Others

尝试了修改lora_target，但依旧没有用