模型融合代码跑不通merge_lora_params.py，求解答

[fengyue20]

执行以下代码： python paddlemix/tools/merge_lora_params.py
--model_name_or_path paddlemix/examples/deepseek_vl2/deepseek-ai/deepseek-vl2-tiny
--lora_path work_dirs/deepseekvl2_tiny_lora_bs16_1e5/checkpoint-60
--merge_model_path paddlemix/tools/merge2

显示 LlavaConfig register success!!!!! LLavaTokenizer register success!!!! [2025-04-03 22:03:17,608] [ INFO] - Loading configuration file paddlemix/examples/deepseek_vl2/deepseek-ai/deepseek-vl2-tiny/config.json Traceback (most recent call last): File "/mnt/storage/jinming/miniconda3/envs/DeepSeek-paddle_fine_tune/lib/python3.10/site-packages/paddlenlp/transformers/auto/configuration.py", line 466, in from_pretrained config_class = CONFIG_MAPPING[config_dict["model_type"]] File "/mnt/storage/jinming/miniconda3/envs/DeepSeek-paddle_fine_tune/lib/python3.10/site-packages/paddlenlp/transformers/auto/configuration.py", line 255, in getitem raise KeyError(key) KeyError: 'deepseek_vl_v2'

During handling of the above exception, another exception occurred:

Traceback (most recent call last): File "/mnt/storage/jinming/djm/deepseekvl2lora/paddle2/PaddleMIX/paddlemix/tools/merge_lora_params.py", line 63, in merge() File "/mnt/storage/jinming/djm/deepseekvl2lora/paddle2/PaddleMIX/paddlemix/tools/merge_lora_params.py", line 41, in merge model_config = AutoConfigMIX.from_pretrained(args.model_name_or_path, dtype=dtype) File "/mnt/storage/jinming/miniconda3/envs/DeepSeek-paddle_fine_tune/lib/python3.10/site-packages/paddlenlp/transformers/auto/configuration.py", line 468, in from_pretrained raise ValueError( ValueError: The checkpoint you are trying to load has model type deepseek_vl_v2 b

求解答

[fengyue20]

当前环境下pip升级了transformer，但是paddle里的Transformer不变，怎么办。微调成功了，但是微调后和原模型融合不了，没法用微调的模型推理啊

[cheng221]

https://github.com/PaddlePaddle/PaddleMIX/pull/1207 新提交的PR，修复LoRA merge 的问题