id string | name string | type string | layer int64 | source string | size string | domain string | license string | link string | paper string | notes string |
|---|---|---|---|---|---|---|---|---|---|---|
AZ-CORP-001 | DOLLMA | corpus | 2 | HuggingFace / allmalab | 651M words | general · news · legal · blogs · books | restricted | https://huggingface.co/datasets/allmalab/DOLLMA | P2 | Primary LLM pretraining corpus; Wiki, news, laws, blogs, books |
AZ-CORP-002 | azWaC | corpus | 2 | SketchEngine | 94M words | web-general | research | https://www.sketchengine.eu/azwac-azerbaijani-corpus/ | P2 | Web-crawled 2012; deduplicated; oldest major AZ corpus |
AZ-CORP-003 | azcorpus_v0 | corpus | 2 | HuggingFace / azcorpus | 1.9M docs | general · politics · science · culture · sport | CC-BY-NC | https://huggingface.co/datasets/azcorpus/azcorpus_v0 | P2 | 1.9M documents across diverse topics for text generation |
AZ-CORP-004 | AZ Wikipedia | corpus | 2 | Wikimedia | ~200K articles | encyclopedic | CC-BY-SA | https://az.wikipedia.org | P2 | Subset used in DOLLMA; also available via Wikimedia dumps |
AZ-CORP-005 | C4-AZ | corpus | 2 | HuggingFace / allenai | 1.83M docs | web-general | Apache-2.0 | https://huggingface.co/datasets/allenai/c4 | P2 | Noisy; significant non-AZ content found on inspection |
AZ-CORP-006 | OSCAR-AZ | corpus | 2 | HuggingFace / oscar-corpus | 316M words | web-general | CC0 | https://huggingface.co/datasets/oscar-corpus/OSCAR-2301 | P2 | Common Crawl filtered subset; used before DOLLMA |
AZ-CORP-007 | AZ Laws corpus | corpus | 2 | allmalab / HuggingFace | ~10M words | legal | open | https://huggingface.co/datasets/allmalab | P2 | Azerbaijani legal texts; independently released by aLLMA Lab |
AZ-CORP-008 | azerbaijani-blogs | corpus | 2 | HuggingFace / aznlp | unknown | blogs | unknown | https://huggingface.co/datasets/aznlp/azerbaijani-blogs | P2 | Blog text corpus; limited documentation |
AZ-CORP-009 | Domrachev-Sudoplatova | corpus | 2 | GitHub | 2.2M words / 100K sents | mixed | research | https://github.com/alexeyev/awesome-azerbaijani-nlp | P2 | Scraped corpus used in morpheme boundary detection research |
AZ-DATA-001 | AZ NER Dataset | dataset | 4 | HuggingFace / LocalDoc | unknown | news · general | CC-BY-NC-ND | https://huggingface.co/datasets/LocalDoc/azerbaijani-ner-dataset | P2·P4 | 25-class NER; annotated Azerbaijani named entities |
AZ-DATA-002 | AZ Sentiment (reviews) | dataset | 4 | HuggingFace / hajili | ~160K reviews | e-commerce | unknown | https://huggingface.co/datasets/hajili/azerbaijani_review_sentiment_classification | P2·P4 | 160K customer reviews with scores; largest AZ sentiment resource |
AZ-DATA-003 | AZ Sentiment (social+news) | dataset | 4 | ACM DL | 50K items | social media · news | research | https://dl.acm.org/doi/10.1145/3661725.3661749 | P4 | 38K comments + 12K news articles; binary sentiment labels |
AZ-DATA-004 | AZ Twitter Sentiment | dataset | 4 | ADA University | ~2K tweets | social media | research | https://www.researchgate.net/publication/339098113 | P4 | First AZ Twitter sentiment dataset; LR/NB/SVM baselines |
AZ-DATA-005 | AZE-SCI | dataset | 4 | HuggingFace / allmalab | unknown | academic · science | restricted | https://huggingface.co/datasets/allmalab | P4 | Text classification dataset; 4+ science categories |
AZ-DATA-006 | AZE-NSP | dataset | 4 | HuggingFace / allmalab | unknown | books | restricted | https://huggingface.co/datasets/allmalab | P4 | Next-sentence prediction; built from Azerbaijani books |
AZ-DATA-007 | CB-MCQ | dataset | 4 | HuggingFace / allmalab | unknown | school · general | restricted | https://huggingface.co/datasets/allmalab | P4 | Closed-book MCQ; middle/high school + driver exam topics |
AZ-DATA-008 | Aze-Instruct-2K | dataset | 2 | HuggingFace / karabakh-nlp | 2001 samples | instruction · reasoning | unknown | https://huggingface.co/karabakh-nlp/AzQ-1.7B | P2 | 12 reasoning domains; used to train AzQ-1.7B |
AZ-DATA-009 | AZ-MRPC (translated) | dataset | 4 | HuggingFace / Eljan Mahammadli | unknown | paraphrase | unknown | https://huggingface.co/datasets | P4 | Azerbaijani translation of MRPC; used in NLU benchmark |
AZ-DATA-010 | UD Azerbaijani-TueCL | dataset | 2 | Universal Dependencies | ~110 sentences | treebank | CC-BY-SA | https://universaldependencies.org | P2·P4 | UD treebank; 20 Cairo + 90 Turkic group sentences |
AZ-BENCH-001 | TUMLU-AZ | benchmark | 4 | GitHub / ceferisbarov | 38139 Qs (8 langs) | school · MMLU | CC-BY | https://github.com/ceferisbarov/TUMLU | P4 | Turkic MMLU; AZ subset spans 11 subjects; natively authored |
AZ-BENCH-002 | TUMLU-mini-AZ | benchmark | 4 | GitHub / ceferisbarov | balanced subset | school · MMLU | CC-BY | https://github.com/ceferisbarov/TUMLU | P4 | Manually verified balanced subset; recommended for evaluation |
AZ-BENCH-003 | aLLMA NLU Benchmark | benchmark | 4 | HuggingFace / allmalab | multi-task | NLU · classification · NSP · QA | restricted | https://arxiv.org/abs/2407.02337 | P4 | First systematic AZ NLU benchmark; covers 4 tasks |
AZ-BENCH-004 | BLEnD-AZ | benchmark | 4 | GitHub / nlee0212 | 52.6K QA pairs (16 countries) | culture · everyday | CC-BY | https://github.com/nlee0212/BLEnD | P4 | Cross-cultural QA; AZ is one of 13 languages |
AZ-MODEL-001 | aLLMA-SMALL | model | 3 | HuggingFace / allmalab | small encoder | general | restricted | https://huggingface.co/allmalab | P3 | Monolingual AZ encoder; trained on DOLLMA; dynamic masking |
AZ-MODEL-002 | aLLMA-BASE | model | 3 | HuggingFace / allmalab | base encoder | general | restricted | https://huggingface.co/allmalab | P3 | Monolingual AZ BERT-base; best AZ-only encoder as of 2024 |
AZ-MODEL-003 | AzQ-1.7B | model | 3 | HuggingFace / karabakh-nlp | 1.7B params | instruction · reasoning | MIT | https://huggingface.co/karabakh-nlp/AzQ-1.7B | P3 | Qwen3-1.7B fine-tuned; 12 reasoning domains; 2025 |
AZ-MODEL-004 | AtLLaMA | model | 3 | HuggingFace / arzuhussein | 8B params | instruction · general | MIT | https://huggingface.co/arzuhussein/atllama | P3 | LLaMA-3.1-8B fine-tuned on 50K AZ instruction pairs |
AZ-MODEL-005 | az-mistral | model | 3 | HuggingFace / interneuronai | 7B params | instruction | MIT | https://huggingface.co/interneuronai/az-mistral | P3 | Mistral-7B fine-tuned on AZ Alpaca; project Barbarossa |
AZ-MODEL-006 | HPLT-BERT-AZ | model | 3 | HuggingFace / HPLT | base encoder | general | Apache-2.0 | https://huggingface.co/HPLT/hplt_bert_base_az | P3 | BERT trained on HPLT web corpus; multilingual project |
AZ-MODEL-007 | mBERT-AZ | model | 3 | HuggingFace / google | multilingual | general | Apache-2.0 | https://huggingface.co/google-bert/bert-base-multilingual-cased | P3 | Baseline multilingual BERT; 104 languages incl. AZ |
AZ-MODEL-008 | XLM-R-AZ | model | 3 | HuggingFace / FacebookAI | multilingual | general | MIT | https://huggingface.co/FacebookAI/xlm-roberta-base | P3 | Strong multilingual baseline; best zero-shot AZ performer |
AZ-MODEL-009 | AZ Sentiment XLM-R | model | 3 | HuggingFace / LocalDoc | multilingual fine-tune | sentiment | CC-BY-NC | https://huggingface.co/LocalDoc/sentiment_analysis_azerbaijani | P3 | XLM-RoBERTa fine-tuned for 3-class AZ sentiment |
AZ-MODEL-010 | AzerBERT (Iranian AZ) | model | 3 | HuggingFace / language-ml-lab | base encoder | general | research | https://huggingface.co/language-ml-lab/iranian-azerbaijani-nlp | P3 | For Iranian Azerbaijani (azb); NER + POS + classification |
AZ-TOOL-001 | Azmorph | tool | 1 | GitHub | morphological analyzer | morphology | research | https://github.com/alexeyev/awesome-azerbaijani-nlp | P1 | Pre-alpha morphological analyzer for Azerbaijani |
AZ-TOOL-002 | MorAz | tool | 1 | EMNLP 2018 | morphological analyzer | morphology | open | https://aclanthology.org/D18-2005/ | P1 | Open-source morphological analyzer; Özenç et al. 2018 |
AZ-TOOL-003 | awesome-azerbaijani-nlp | tool | 1 | GitHub / alexeyev | resource list | meta | MIT | https://github.com/alexeyev/awesome-azerbaijani-nlp | P1 | Curated list of AZ NLP resources; precursor to ALMAZ roster |
YAML Metadata Warning: empty or missing yaml metadata in repo card
Check out the documentation for more information.
The ALMAZ Resource Roster — a curated catalog of Azerbaijani NLP artifacts. 36 artifacts covering corpora, datasets, models, benchmarks, and tools.
GitHub: https://github.com/almaz-nlp/almaz-roster DOI: https://doi.org/10.5281/zenodo.19023843
- Downloads last month
- 3