Datasets:

Modalities:
Text
Formats:
arrow
ArXiv:
Libraries:
Datasets
License:
Shaltiel's picture
Upload folder using huggingface_hub
950ceac verified
metadata
license: odc-by
language:
  - he
  - en
pretty_name: DictaLM2.0 Quantization Calibration Dataset

DictaLM 2.0 Quantization Calibration Dataset

This dataset consists of documents which can be used as a calibration dataset for quantizing Hebrew-English LLMs, such as DictaLM 2.0. You can read more about the model on the release page or in the technical report.

The dataset consists of 1024 documents - 512 in English (FineWeb-EDU) and 512 in Hebrew (Wikipedia, OSCAR).