You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Бұл репозиторийге қолжетімділік өтінім бойынша беріледі. Өзіңіз туралы және деректерді қалай қолданатыңыз туралы жазыңыз — өтінімді TilQazyna командасы қарайды. · Доступ к репозиторию выдаётся по заявке. Расскажите о себе и о том, как собираетесь использовать данные — заявку рассматривает команда TilQazyna. · Access to this repository is granted on request. Tell us who you are and how you plan to use the material; the TilQazyna team reviews each application.

Log in or Sign Up to review the conditions and access this dataset content.

Til-Instruct

Нұсқау бойынша жауап беруге арналған жинақ · Датасет инструкций и ответов · Instruction-following dataset

Қазақша · Русский · English


Қазақша

Til-Instruct — тілдік модельдерді нұсқау орындауға бейімдеуге арналған 2 563 966 жазбалық жинақ. Көлемі — 8.35 ГБ; қазақ және ағылшын тілдеріндегі жазбалар басым.

Құрамы мен деңгейлері

Әр жазбада messages, task, source, src_lang, score, category және lang өрістері сақталған. Модель-судья жауаптарды бағалап, жинақты жүктеу конфигтері бойынша бөлген.

Конфиг Жол саны
raw 2 563 966
clean 2 114 016
premium 1 581 787
Тіл Жол саны
en 1 298 253
kk 1 123 543
other 67 956
mixed 58 296
ru 15 918

Дереккөз бойынша stk үлесіне 2 388 894, ал tq үлесіне 175 072 жазба тиесілі. Мұндай құрылым тіл, міндет және сапа қабаты бойынша іріктеу жасауға қолайлы.

Жүктеу

from datasets import load_dataset

ds = load_dataset(
    "TilQazyna/Til-Instruct",
    "premium",
    split="train",
    streaming=True,
)
print(next(iter(ds)))

streaming=True параметрі жазбаларды толық жинақты алдын ала көшірмей оқуға мүмкіндік береді.

Қолжетімділік

Репозиторий карточкасы мен файлдар тізімі бәріне көрінеді. Деректерді жүктеу үшін «Request access» арқылы өтінім жіберіледі, оны TilQazyna командасы қарайды.

Байланысты репозиторийлер

Жинақ til-instruct-mix-v1 және TilQazyna тапсырмалық деректерінен құралған. Онымен Til-Core-0.5B-Instruct, Til-Core-1B-Instruct және Til-2B-instruct модельдері оқытылады.


Русский

Til-Instruct — набор из 2 563 966 записей для обучения языковых моделей следовать инструкциям. Объём — 8.35 ГБ; основную часть составляют записи на казахском и английском языках.

Состав и уровни

Каждая запись содержит поля messages, task, source, src_lang, score, category и lang. Модель-судья оценила ответы, после чего записи распределили по конфигам загрузки.

Конфиг Строк
raw 2 563 966
clean 2 114 016
premium 1 581 787
Язык Строк
en 1 298 253
kk 1 123 543
other 67 956
mixed 58 296
ru 15 918

Источник stk дал 2 388 894 записи, источник tq — 175 072 записи.

Как загрузить

from datasets import load_dataset

ds = load_dataset(
    "TilQazyna/Til-Instruct",
    "premium",
    split="train",
    streaming=True,
)
print(next(iter(ds)))

Параметр streaming=True читает записи без предварительной загрузки всего набора.

Доступ

Карточка репозитория и перечень файлов открыты всем. Скачивание начинается после одобрения заявки, поданной через «Request access»; её рассматривает команда TilQazyna.

Связанные репозитории

В состав входит til-instruct-mix-v1 и наборы TilQazyna для отдельных задач. На этих данных обучаются Til-Core-0.5B-Instruct, Til-Core-1B-Instruct и Til-2B-instruct.


English

Til-Instruct contains 2,563,966 records for instruction-tuning language models. It occupies 8.35 GB, with Kazakh and English accounting for most records.

Contents and tiers

Each record has messages, task, source, src_lang, score, category, and lang fields. A judge model scored the responses, and the resulting tiers are exposed as loading configs.

Config Rows
raw 2,563,966
clean 2,114,016
premium 1,581,787
Language Rows
en 1,298,253
kk 1,123,543
other 67,956
mixed 58,296
ru 15,918

The stk source contributes 2,388,894 records, while tq contributes 175,072 records.

Loading the data

from datasets import load_dataset

ds = load_dataset(
    "TilQazyna/Til-Instruct",
    "premium",
    split="train",
    streaming=True,
)
print(next(iter(ds)))

With streaming=True, records can be consumed without downloading the complete dataset first.

Access

The repository card and file list remain visible to everyone. To download the data, use “Request access”; the TilQazyna team reviews each submission before granting access.

Related repositories

The collection draws on til-instruct-mix-v1 and TilQazyna task datasets. It feeds Til-Core-0.5B-Instruct, Til-Core-1B-Instruct, and Til-2B-instruct.


Лицензия · License: apache-2.0 · TilQazyna

Downloads last month
6

Collection including TilQazyna/Til-Instruct