Which folder should I use to train model?
I see two datasets, namely emb_ Recall and keyword_ Recall, should I use both two folder datasets for training, right?
I see two datasets, namely emb_ Recall and keyword_ Recall, should I use both two folder datasets for training, right?
Both are OK. emb_recall is plain text, keyword_recall is multiple-choice questions.
This dataset works well! Do we have plans for larger dataset? I wanna improve more MMLU score.
This dataset works well! Do we have plans for larger dataset? I wanna improve more MMLU score.
In the dataset under consideration, the component labeled as 'emb_recall' constitutes the knowledge base, while the 'keyword_recall' section functions as instructions. Significant enhancements through MMLU are primarily attributed to the alignment of pre-existing model knowledge with the instructions provided in the 'keyword_recall' section, rather than an augmentation of the model's knowledge base. Consequently, this alignment fails to elevate the model's perceptible performance. To authentically augment the model with MMLU-specific knowledge, a focus on expanding the training with additional knowledge-centric content is required, as opposed to merely focusing on the alignment of instructions.