pinkelephantlimited/

pinkelephant-llm-48b-sft-tokenized

The instruction-tuning corpus behind Pink Elephant 48B — pre-tokenized and ready to train. This dataset contains 99,661 training examples and 1,000 evaluation examples used to fine-tune the Pink Elephant 48B MoE model. All examples are already tokenized with the Pink Elephant tokenizer (vocabulary 100,352), so training can begin with no preprocessing.

	Schema




	

Column Type Description

input_ids List[int32]… See the full description on the dataset page: https://huggingface.co/datasets/pinkelephantlimited/pinkelephant-llm-48b-sft-tokenized.