# Low-resource language optimization

This is a version of the NLLB-1.3b model fine-tuned with parallel data for 29 Finno-Ugric languages, supporting the generation of multiple dialects/variants.

This is a Bambara automatic speech recognition model fine-tuned based on the NVIDIA NeMo framework, suitable for Bambara speech-to-text tasks.

A Ukrainian-specific speech synthesis model fine-tuned based on Fish Speech 1.5, supporting high-quality voice generation for 55 speakers

Uzbek automatic speech recognition model fine-tuned from OpenAI Whisper-small on Common Voice 17.0 dataset

Llama-SEA-LION-v3-8B is a multilingual large language model optimized for Southeast Asian languages, supporting 11 Southeast Asian languages and having undergone continuous pre-training on approximately 200 billion tokens.

Fish Speech V1.5 is a leading text-to-speech (TTS) model trained on over 1 million hours of multilingual audio data.

NLLB-200 is a neural machine translation model supporting 200 languages, focusing on translation research for low-resource languages.

OpenLID-v2 is a high-coverage, high-performance language identification model supporting 200 language variants, an improved version of OpenLID.

EXLMR is an extended version of XLM-R that supports new languages by expanding the tokenizer vocabulary to mitigate out-of-vocabulary issues, specifically optimized for low-resource Ethiopian languages.

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the IMaSC, Indic TTS Malayalam, and OpenSLR Malayalam training datasets, supporting automatic speech recognition for Malayalam.

A self-supervised speech model based on the HuBERT architecture, specifically optimized for 21 languages in Sub-Saharan Africa with 60,000 hours of training data

The Poro 34B Chat version is an instruction-following model fine-tuned based on Poro 34B, supporting bilingual interaction in Finnish and English, jointly developed by Silo AI, the TurkuNLP group, and HPLT.

Viking 33B is a 33 billion parameter decoder-only Transformer model supporting Finnish, English, and multiple Nordic languages, with capabilities in code understanding and generation.

Viking 13B is a 13-billion-parameter multilingual large model supporting Finnish, English, and other Nordic languages, with code processing capabilities

Viking-7B is a 7-billion-parameter Transformer model specializing in Finnish, Nordic languages, and programming code, trained on 2 trillion tokens.

NLLB-MoE is a Mixture of Experts machine translation model developed by Meta, supporting 200 languages, and is one of the most advanced open-access machine translation models available.

GPT-SW3 is a large-scale Nordic language model developed by AI Sweden, supporting text generation tasks in 5 Nordic languages and English.

A general-purpose language model supporting over 100 languages, suitable for various natural language processing tasks.

A multilingual processing model supporting over 100 languages, suitable for various natural language processing tasks.

A multilingual processing model supporting over 200 languages, suitable for various natural language processing tasks.

A wav2vec2-based speech recognition model, fine-tuned at the phoneme level for Amharic

NLLB-CLIP is a multilingual vision-language model combining the NLLB model's text encoder with CLIP's image encoder, supporting 201 languages.

NLLB-CLIP is a multilingual vision-language model combining the NLLB text encoder with the CLIP image encoder, supporting 201 languages

Samoan text-to-speech model developed by Meta, based on the VITS architecture, supporting high-quality speech synthesis

Creole text-to-speech model developed by Meta, utilizing VITS end-to-end architecture for high-quality speech synthesis

A Kazakh language model based on mGPT-XL (1.3B) with 1.3 billion parameters, specifically optimized for Kazakh

A multilingual processing model supporting over 100 languages and scripts, covering from mainstream languages to various dialects and regional variants.

A speech recognition model fine-tuned on Sinhala language based on OpenAI Whisper-small

A GPT-2 language model trained from scratch on Kazakh news and Wikipedia corpus, supporting Latin script Kazakh text generation and processing.

NLLB-Mo is a multilingual machine translation model that supports translation tasks for over 200 languages.

NLLB-200 distilled version with 600M parameters, supporting machine translation tasks for 200 languages

SMaLL-100 is a compact and fast large-scale multilingual machine translation model, covering over 10,000 language pairs, with performance comparable to M2M-100 but smaller in size and faster.

The first neural machine translation system for the Erzya language, improved based on the LaBSE model, supporting sentence embedding representations for Russian and Erzya

A multilingual processing model supporting over 100 languages and writing systems

This is an automatic speech recognition model for Wolof, based on the wav2vec 2.0 architecture, trained on the DVoice dataset, supporting Wolof speech transcription.

This model is a Finnish speech recognition model fine-tuned from facebook/wav2vec2-xls-r-300m on a general speech dataset.

This model is an automatic speech recognition system trained on the DVoice Dari dataset, using the wav2vec 2.0 architecture, supporting transcription of Moroccan Arabic dialect.

BLOOM is a multilingual large language model developed by BigScience, supporting 46 natural languages and 12 programming languages, built on the Transformer architecture with 560 million parameters.

Text summarization model fine-tuned on Arabic wiki_lingua dataset based on mbart-large-50

This is a neural machine translation model for translating from English to North Germanic languages (including Danish, Faroese, Icelandic, Norwegian Bokmål, Norwegian Nynorsk, and Swedish), part of the OPUS-MT project.