Datasetpublic

erenyeager-1/tiny-aya-l2-thinker-multilingual-reasoning

Tiny Aya L2 Multilingual Reasoning (44 languages) Translated multilingual reasoning traces used to train Tiny Aya L2-Thinker. Each example has the prompt, thinking, and answer in the same non-English language alongside the original texts in English. Languages (44) Language Train Test Total Amharic (am) 3,807 448 4,255 Arabic (ar) 22,968 2,538 25,506 Bulgarian (bg) 4,177 452 4,629 Bengali (bn) 3,803 422 4,225 Catalan (ca) 4,251 512 4,763 Czech… See the full description on the dataset page: https://huggingface.co/datasets/erenyeager-1/tiny-aya-l2-thinker-multilingual-reasoning.

sourceHugging Facecc-by-nc-4.0updated 14d agoView on Hugging Face

0likes286downloads

Dataset Card

Tiny Aya L2 Multilingual Reasoning (44 languages)

Translated multilingual reasoning traces used to train **Tiny Aya L2-Thinker**. Each example has the prompt, thinking, and answer in the same non-English language alongside the original texts in English.

Languages (44)

Language	Train	Test	Total
Amharic (`am`)	3,807	448	4,255
Arabic (`ar`)	22,968	2,538	25,506
Bulgarian (`bg`)	4,177	452	4,629
Bengali (`bn`)	3,803	422	4,225
Catalan (`ca`)	4,251	512	4,763
Czech (`cs`)	4,410	457	4,867
German (`de`)	20,438	2,306	22,744
Greek (`el`)	4,423	489	4,912
Basque (`eu`)	3,280	365	3,645
Persian (`fa`)	4,087	430	4,517
Finnish (`fi`)	3,334	412	3,746
Filipino (`fil`)	2,852	336	3,188
French (`fr`)	21,538	2,317	23,855
Irish (`ga`)	3,967	470	4,437
Hausa (`ha`)	4,552	499	5,051
Hebrew (`he`)	4,036	453	4,489
Hindi (`hi`)	4,084	478	4,562
Hungarian (`hu`)	3,622	427	4,049
Indonesian (`id`)	4,435	472	4,907
Igbo (`ig`)	4,632	542	5,174
Italian (`it`)	4,143	470	4,613
Japanese (`ja`)	20,375	2,349	22,724
Javanese (`jv`)	5,198	601	5,799
Khmer (`km`)	3,420	421	3,841
Korean (`ko`)	18,999	2,036	21,035
Lithuanian (`lt`)	3,894	414	4,308
Malay (`ms`)	3,198	367	3,565
Maltese (`mt`)	2,851	342	3,193
Norwegian (`no`)	4,513	504	5,017
Punjabi (`pa`)	3,520	371	3,891
Polish (`pl`)	2,620	293	2,913
Russian (`ru`)	4,649	495	5,144
Slovak (`sk`)	4,292	472	4,764
Swahili (`sw`)	4,445	479	4,924
Tamil (`ta`)	3,403	415	3,818
Telugu (`te`)	3,462	411	3,873
Thai (`th`)	4,085	451	4,536
Turkish (`tr`)	3,426	407	3,833
Ukrainian (`uk`)	4,158	462	4,620
Urdu (`ur`)	3,291	350	3,641
Vietnamese (`vi`)	4,611	508	5,119
Yoruba (`yo`)	4,023	443	4,466
Chinese (`zh`)	4,329	463	4,792
Zulu (`zu`)	3,991	447	4,438
Total			286,388

Load

python

from datasets import concatenate_datasets, get_dataset_config_names, load_dataset

# one language
ds = load_dataset("CohereLabs/tiny-aya-l2-thinker-multilingual-reasoning", "ar")

# every language
configs = get_dataset_config_names("CohereLabs/tiny-aya-l2-thinker-multilingual-reasoning")
all_langs = concatenate_datasets(
    [load_dataset("CohereLabs/tiny-aya-l2-thinker-multilingual-reasoning", name, split="train") for name in configs]
)

Columns

—id: {language_code}-{domain}-{index} (for example ar-math-000001)
—language, language_code: full name and ISO code
—domain: math, science, or general
—question, thinking, output in the given language
—original_question, original_thinking, original_output: English source