metadata
dataset_info:
features:
- name: prompt
dtype: string
- name: completion
dtype: string
splits:
- name: train
num_bytes: 10982258961
num_examples: 2320284
download_size: 2664867981
dataset_size: 10982258961
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
An instruction dataset for speech->text and text->speech.
This speech data is tokenized using the SpeechTokenize approach: https://arxiv.org/abs/2308.16692.
You can do standard finetuning on this dataset using any LLM!