This dataset contains 100k samples (90k train, 10k val) mixing:

50k longhealth samples 50k longcode samples

import msgpack from huggingface_hub import hf_hub_download

filepath = hf_hub_download( repo_id="agu18dec/longhealth-longcode-50k", filename="mixture.msgpack", repo_type="dataset" )

with open(filepath, 'rb') as f: data = msgpack.unpackb(f.read(), raw=False)… See the full description on the dataset page: https://huggingface.co/datasets/agu18dec/longhealth-longcode-50k.