Hello, thank you for the great work. I have the following question.
For the fine-tuning stage where you used the following datasets (please correct me if I am wrong):
Quora-Arabic, Code-Alpaca-Arabic, Alpaca-Arabic, Evol-Instruct-Arabic, and ShareGPT (English original version 80k samples). These datasets are also open-sourced by you on your official HuggingFace repository.
Which training pipeline/ codebase you used for the Localized Supervised Fine-Tuning on the above datasets? I am using FastChat repository to reproduce the results.
Also, can you share the results of Localized Supervised Finetuning and after RLAI for ArabicMMLU and Exams? Table 9 only shows the results for base models and not sure if it's before RLAI or after RLAI. can you please provide some clarification on it?
Hello, thank you for the great work. I have the following question.
For the fine-tuning stage where you used the following datasets (please correct me if I am wrong):
Quora-Arabic, Code-Alpaca-Arabic, Alpaca-Arabic, Evol-Instruct-Arabic, and ShareGPT (English original version 80k samples). These datasets are also open-sourced by you on your official HuggingFace repository.
Which training pipeline/ codebase you used for the Localized Supervised Fine-Tuning on the above datasets? I am using FastChat repository to reproduce the results.
Also, can you share the results of Localized Supervised Finetuning and after RLAI for ArabicMMLU and Exams? Table 9 only shows the results for base models and not sure if it's before RLAI or after RLAI. can you please provide some clarification on it?