关于报错UserWarning: Could not find a config file in MiniCPM-V-2 - will assume that the vocabulary was not modified.

[daihuidai]

是否已有关于该错误的issue或讨论？ | Is there an existing issue / discussion for this?

• [X] 我已经搜索过已有的issues和讨论 | I have searched the existing issues / discussions

该问题是否在FAQ中有解答？ | Is there an existing answer for this in FAQ?

• [x] 我已经搜索过FAQ | I have searched FAQ

当前行为 | Current Behavior

你好，我在微调过程中前几次当验证完进行eval时出现了UserWarning: Could not find a config file in MiniCPM-V-2 - will assume that the vocabulary was not modified.这个报错，我查看了huggingface的issue中该问题的回答，回答说可以忽略，并且模型正常保存和验证，实际上我这边也是正常保存。但是我在验证时，模型的回答和我设置的标签相差较大，请问这个警告是否可以忽略。

下面是我的推理代码请问是否正确呢：

期望行为 | Expected Behavior

No response

复现方法 | Steps To Reproduce

No response

运行环境 | Environment

- OS:
- Python:
- Transformers:
- PyTorch:
- CUDA (`python -c 'import torch; print(torch.version.cuda)'`):

备注 | Anything else?

No response

[LDLINGLINGLING]

这个标签一般是可以忽略的 回答的差别大，是因为你没有正确的加载lora模型。

from peft import AutoPeftModelForCausalLM import torch import os from transformers import AutoTokenizer

path_to_adapter="/root/ld/ld_project/MiniCPM-V/finetune/output/output_minicpmv2_lora/checkpoint-10"

merge_path="/root/ld/ld_project/MiniCPM-V/finetune/output/output_minicpmv2_lora/merge"

if not os.path.exists(merge_path): os.makedirs(merge_path)

model = AutoPeftModelForCausalLM.from_pretrained(

path to the output directory

path_to_adapter,
device_map="auto",
trust_remote_code=True

).eval()

vpm_resampler_embedtokens_weight = torch.load(f"{path_to_adapter}/vpm_resampler_embedtokens.pt")

model.load_state_dict(vpm_resampler_embedtokens_weight, strict=False) #此时模型已经可用，加载了vit的resample和llm的lora