How is the dataset ued to train?

[Pisces032]

I'm trying to use PEFT to improve the model. I wonder how AnghaBench_compile.jsonl is used to train. i noticed declare -a dataset=( "path_to_llm4decompile_data/arrow/part-00000" ) in run_llm4decompile_train.sh, but i can't make out the training process. Maybe colossalai format hides some details about the model or the training process? Thank you so much!

[rocky-lq]

Thanks for your interest in our project, we've updated the guidance for preparing Colossal AI training data. Please refer to Prepare the data.

Additionally, we recommend using LLaMA Factory to train the llm4decompile model, as it is more user-friendly. For more details, please visit LLaMA-Factory.

[Pisces032]

Thank you!