datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
libretranslate-en-kab-suggestions
Kabyle Suggestions Dataset
This dataset contains English-to-Kabyle translation suggestions submmitted by users using LibreTranslate, designed to support the development and evaluation of machine translation tools for the Kabyle language.
sap_faqsf_newmanas-dataset-v2
Manas Dataset
Statistics
Total clean conversations: 1122
Train: 1009
Eval: 113
Format
{
"conversations": [
{"from": "system", "value": "..."},
{"from": "human", "value": "..."},
{"from": "gpt", "value": "..."}
]
}
KabPunct
KabPunct
A punctuation and capitalisation restoration corpus for Kabyle (Taqbaylit, kab, Latin
script), from the AƔBALU project.
1,318,707 word-labelled sentences drawn from the full AƔBALU-Text v1 corpus and
speaker-disjoint Common Voice Kabyle splits. Every sentence is broken into lowercased ASR
tokens with two parallel label sequences: which punctuation mark follows each word, and how
the word is capitalised. It is the training and evaluation corpus for
agbalu/Belaid-31M… See the full description on the dataset page: https://huggingface.co/datasets/agbalu/KabPunct.KabG2P
KabG2P
A grapheme-to-phoneme pronunciation dictionary for Kabyle (Taqbaylit, kab, Latin script),
from the AƔBALU project.
25,634 Kabyle word–IPA pairs recovered by aligning 292,921 tokens across 59,462
sentence pairs at a 99.53% alignment rate, with a 0% ambiguity rate across the entire
vocabulary. Every attested word has exactly one IPA reading. It is the phonetics layer
underlying agbalu/Matoub-82M and
agbalu/Fadhma-300M, and the reference
target for any Kabyle G2P model.
from… See the full description on the dataset page: https://huggingface.co/datasets/agbalu/KabG2P.KabLex
KabLex
A lexical layer for Kabyle (Taqbaylit, kab, Latin script), from the
AƔBALU project.
366,892 lexical entries merged from three permissively licensed sources and normalised to
one orthography, plus 25,642 word–pronunciation pairs aligned from sentence-level
grapheme-to-phoneme data.
from datasets import load_dataset
lex = load_dataset("agbalu/KabLex", "lexicon", split="train")
ipa = load_dataset("agbalu/KabLex", "pronunciations", split="train")
lexicon
366… See the full description on the dataset page: https://huggingface.co/datasets/agbalu/KabLex.KabStandard
KabStandard
A 497,944-pair parallel dataset for Kabyle orthography standardisation — mapping informal,
French-keyboard and Arabizi Kabyle text to canonical Kabyle Latin orthography. Derived from
the Latin side of agbalu/KabTifinagh
by a deterministic seeded probabilistic corruption pass that simulates the keyboard strategies
Kabyle speakers use on phones and social media.
Used to train agbalu/Boulifa-48M, which reaches
97.39% character accuracy on held-out test pairs under… See the full description on the dataset page: https://huggingface.co/datasets/agbalu/KabStandard.SAP_JKabyle_Road_Traffic_Code
Kabyle-English Road Traffic Code Dataset
A bilingual parallel corpus of 102 road traffic signs and regulations in English and Kabyle (Taqbaylit), an Amazigh language spoken in Algeria.
Categories
Dangers (Imihiten): Warning signs (39 entries)
Prohibitions (Tigedlin): Prohibitory signs (35 entries)
Obligations (Timariwin): Mandatory signs (16 entries)
End of Restrictions: End of regulation signs (12 entries)
Splits
Split
Size
Train
62… See the full description on the dataset page: https://huggingface.co/datasets/taqbaylit/Kabyle_Road_Traffic_Code.KabBench
KabBench
Evaluation data for Kabyle (Taqbaylit, kab, Latin script), from the
AƔBALU project.
Two configs. mt is a repaired Kabyle reference for machine translation — the public one
is 16.2% corrupt. lid is a balanced six-language set for telling Kabyle apart from its
Berber siblings, which the identifiers in common use cannot do.
from datasets import load_dataset
mt = load_dataset("agbalu/KabBench", "mt", split="devtest")
lid = load_dataset("agbalu/KabBench", "lid"… See the full description on the dataset page: https://huggingface.co/datasets/agbalu/KabBench.rustbench_testwesnoth-ethea-canon-campaignsfootball-culture-reasoning
Football Culture Reasoning Bench v1
Expert-graded evaluation of LLM reasoning about football (soccer) fandom culture: subcultural concepts, non-Western specificity (Japan / J.League, South America, Asia), macro-sociological context, and stereotype avoidance.
This is a small, deliberately hard proof set (20 items). The failures it probes are cultural, not linguistic — candidate answers are fluent and confident, but stale, West-centric, or normatively preachy.
Task… See the full description on the dataset page: https://huggingface.co/datasets/Kablue/football-culture-reasoning.medical-alpacatrh1kabyle-wikipedia
Kabyle Wikipedia Dataset
This dataset contains articles from the Kabyle (Tamazight) Wikipedia, scraped for LLM training.
Dataset Structure
title: Article title
url: Article URL
text: Cleaned text content
Collection Method
Scraped using Scrapling on Modal.com.
pashto-kabul-treaty-1921-sft
Dataset Card for Pashto Kabul Treaty 1921 SFT
Dataset Summary
This dataset contains the complete Pashto translation of the 1921 Treaty between the British and Afghan Governments (also known as the Kabul Treaty), along with 100 question-answer pairs derived from the treaty text. The original treaty was signed at Kabul on November 22, 1921, and ratifications were exchanged on February 6, 1922.
The dataset is designed for Supervised Fine-Tuning (SFT) of Large… See the full description on the dataset page: https://huggingface.co/datasets/nassimjp/pashto-kabul-treaty-1921-sft.english-kabyle-parallel
English – Kabyle Parallel Corpus (Tatoeba)
130 883 aligned sentence pairs extracted from the open Tatoeba database.
Pipeline
Download – raw Tatoeba dumps
Gather – filter English & Kabyle sentences
Align – pair by sentence-id
Fix – normalise
All steps were performed with the kabyle-nlp-toolkit.
File
File
Lines
Size
Format
en-kab-parallel.jsonl
130 883
10.7 MiB
One JSON object per line: {"en": "…", "kab": "…"}
Example
{"en":… See the full description on the dataset page: https://huggingface.co/datasets/Imsidag-community/english-kabyle-parallel.gemini-3-pro-preview-high-reasoning-1000xThis is a reasoning dataset created using Gemini 3 Pro Preview with a reasoning depth set to high. Some of these questions are from reedmayhew and the rest were generated.
The dataset is meant for creating distilled versions of Gemini 3 Pro Preview by fine-tuning already existing open-source LLMs on the summarized reasoning traces provided from their API.
This dataset includes 250x from TeichAI/gemini-3-pro-preview-high-reasoning-250x
Stats
Costs: $ 32.7 (USD)
Tokens: 2.73 M… See the full description on the dataset page: https://huggingface.co/datasets/kabsis/gemini-3-pro-preview-high-reasoning-1000x.kab-en-toponyms-sentences
English-Kabyle Parallel Corpus for Machine Translation
This dataset contains 32,024 grammatically flawless parallel sentence pairs mapping English to literary Kabyle (Taqbaylit kab).
This corpus was synthesized using a linguistically-informed morphosyntactic rule engine paired with clean OpenStreetMap toponym registries from boffire/kabyle-toponyms. It handles complex phonetic mutations natively, making it a state-of-the-art bootstrapping asset for fine-tuning… See the full description on the dataset page: https://huggingface.co/datasets/taqbaylit/kab-en-toponyms-sentences.kabyle-corpussol_tvgemini-3-flash-preview-1000x
Gemini 3 Flash Preview - 1,000x
This is a reasoning dataset created using Gemini 3 Flash Preview with a reasoning depth set to high.
The dataset is meant for creating distilled versions of Gemini 3 Flash Preview by fine-tuning already existing open-source LLMs.
Some of these prompts are from reedmayhew, the rest were generated
Stats
Costs: $ 2.96 (USD)
Total tokens (input + output): 1.01 M
Generated using DataGen by TeichAI
claude-sonnet-4.5-high-reasoning-250xThis is a reasoning dataset created using Claude Sonnet 4.5 with a high reasoning effort. Some of these questions are from reedmayhew and the rest were generated.
The dataset is meant for creating distilled versions of Claude Sonnet 4.5 by fine-tuning already existing open-source LLMs.
The default system prompt from OpenrouterAI was used
You are Claude Sonnet 4.5, a large language model from anthropic.
Formatting Rules:
- Use Markdown for lists, tables, and styling.
- Use ```code fences```… See the full description on the dataset page: https://huggingface.co/datasets/kabsis/claude-sonnet-4.5-high-reasoning-250x.gpt-4o-traces
Dataset Card for Dataset Name
Dataset Details
Dataset Description
Curated by: [More Information Needed]
Funded by [optional]: [More Information Needed]
Shared by [optional]: [More Information Needed]
Language(s) (NLP): [More Information Needed]
License: [More Information Needed]
Dataset Sources [optional]
Repository: [More Information Needed]
Paper [optional]: [More Information Needed]
Demo [optional]: [More Information Needed]… See the full description on the dataset page: https://huggingface.co/datasets/kabelomalapane/gpt-4o-traces.graph1combinedkabel1.json