Datasets:

Modalities:
Text
Formats:
arrow
ArXiv:
Libraries:
Datasets
License:
File size: 594 Bytes
950ceac
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
---
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](https://huggingface.co/dicta-il/dictalm2.0). You can read more about the model on the [release page](https://dicta.org.il/dicta-lm) or in the [technical report](https://arxiv.org/abs/2407.07080).

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