"trust_remote_code=True" problem

[yingxin-chen]

Hi author, when I use your project to fine-tune a local model, I get the following prompt
"trust_remote_code=True"
Where should I change the code? Can you provide an example?

[research4pan]

Thanks for your interest in LMFlow! You may set --trust_remote_code 1 when running the python script or deepspeed script

deepspeed ${deepspeed_args} \
  examples/finetune.py \
    --trust_remote_code 1   \
    --model_name_or_path ${model_name_or_path} \
    --dataset_path ${dataset_path} \
    --output_dir ${output_dir} --overwrite_output_dir \
    --num_train_epochs 0.01 \
    ...
    | tee ${log_dir}/train.log \
    2> ${log_dir}/train.err

or

python3 \
  examples/finetune.py \
    --trust_remote_code 1   \
    --model_name_or_path ${model_name_or_path} \
    --dataset_path ${dataset_path} \
    --output_dir ${output_dir} --overwrite_output_dir \
    --num_train_epochs 0.01 \
    ...
    | tee ${log_dir}/train.log \
    2> ${log_dir}/train.err

Or if you prefer to modify the python code directly, you may check the code pieces located in src/lmflow/models/hf_decoder_model.py. There are several other places, which you can modify them all to allow trust_remote_code to be switched on for both inference and finetune. Hope that can be helpful 😄