Skip to content

[Bug] NameError: name 'GPTModel' is not defined during evaluation in SFT LoRA training #3759

@peteli25

Description

@peteli25

请提出你的问题

在使用 PaddleFormers 进行 SFT LoRA 微调时,训练过程在 evaluation 阶段报错:NameError: name 'GPTModel' is not defined

使用以下配置进行 LoRA SFT 训练:

model_name_or_path: baidu/ERNIE-4.5-0.3B-Base-PT
lora: true
lora_rank: 8
stage: SFT
fine_tuning: lora
do_train: true
do_eval: true
eval_steps: 100
evaluation_strategy: steps

Traceback (most recent call last):
File "/home/aistudio/PaddleFormers/paddleformers/cli/launcher.py", line 40, in
launch()
...
File "/home/aistudio/PaddleFormers/paddleformers/trainer/trainer.py", line 4442, in evaluation_loop
or (isinstance(self.model, LoRAModel) and isinstance(self.model.model, GPTModel))
NameError: name 'GPTModel' is not defined

Metadata

Metadata

Assignees

Labels

questionFurther information is requested

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions