metadata
license: mit
dataset_info:
features:
- name: conversations
list:
- name: from
dtype: string
- name: value
dtype: string
splits:
- name: train
num_bytes: 511907485
num_examples: 359706
download_size: 233494201
dataset_size: 511907485
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
task_categories:
- text-generation
- text-classification
- question-answering
language:
- uz
tags:
- orca
- code
- art
- music
- legal
- finance
- biology
- chemistry
pretty_name: SlimOrca-Dedup Uzbek
size_categories:
- 100K<n<1M
This is an Uzbek translated and cleaned version of https://huggingface.co/datasets/Open-Orca/SlimOrca-Dedup.
Specifically, these replaced/removed records that had 'Uzbek translation|Uzbekcha tarjima|Uzbek tarjima|impossible to translate|not possible to translate|cannot fulfill your request|text is in|tilida yozilgan|Uzbek|o'zbek|ozbek|I am sorry'.
You can use this dataset for chat fine-tuning of LLMs.
This dataset has around 100M tokens (500M*0.8/4 = 100M assuming 4 chars are one token).
Check here for uncleaned English to Uzbek pairs version: https://huggingface.co/datasets/MLDataScientist/SlimOrca-Dedup-English-Uzbek