# Low-resource language support

A multilingual audio-to-text model based on Llama-3.1-8B-Instruct, supporting processing of over 40 languages

Latxa 3.1 70B Instruct is an instruction-tuned version based on Llama-3.1 (Instruct), specifically optimized for Basque language, demonstrating excellent performance on multiple Basque benchmarks.

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

A fine-tuned mBART model for translating Moroccan dialect (Darija) to English, suitable for daily conversations and informal text translation.

ADIA_TTS is an open-source Wolof speech synthesis model developed by CONCREE, based on the parler-tts-mini-multilingual-v1.1 model, achieving significant progress in Wolof speech synthesis.

Jina Embeddings V3 is a multilingual sentence embedding model supporting over 100 languages, focusing on sentence similarity calculation and feature extraction tasks.

Jina Embeddings V3 is a multilingual sentence embedding model supporting over 100 languages, focusing on sentence similarity calculation and feature extraction tasks.

Turkish large language model based on Gemma-2-9b developed by WiroAI, specializing in dialogue generation tasks

This model is fine-tuned based on mbart-large-50-many-to-many-mmt, specifically designed for converting Romanized Bengali to standard Bengali script

kazRush-kk-ru is a Kazakh-to-Russian translation model based on the T5 configuration, trained on multiple parallel datasets.

Pathumma-llm-audio-1.0.0 is an 8-billion-parameter Thai large language model specifically designed for audio comprehension tasks, capable of processing various audio inputs including speech, general audio, and music.

A large Bengali language model optimized based on the Llama-3.2-3B architecture, with 42K Bengali tokens extended and fine-tuned, performing excellently in Bengali understanding and generation tasks.

A Hindi speech synthesis model fine-tuned on the common_voice_17_0 dataset based on microsoft/speecht5_tts

This is a translation model with 418 million parameters, specifically built for translating English into Yoruba, Igbo, and Hausa.

The GTE multilingual base model is a dense sentence transformer that supports sentence similarity calculation and text embedding tasks in multiple languages.

Mistral-7B instruction fine-tuned model optimized for Uzbek, supporting various NLP tasks in Uzbek and English

EuroLLM-1.7B-Instruct is the first instruction fine-tuned model in the EuroLLM series. It has multilingual processing capabilities and can understand and generate text in multiple European and other related languages. It performs excellently in tasks such as machine translation.

UCCIX-Llama2-13B-Instruct is an Irish-English bilingual large language model, developed based on the Llama 2-13B architecture, with special optimizations for Irish language processing.

This is a Shona text-to-speech model fine-tuned based on SpeechT5, capable of converting Shona text into natural speech.

This is a Transformer-based text-to-speech (TTS) model that supports Dhivehi language speech synthesis.

A translation model supporting bidirectional translation between Serbian and English

A multilingual machine translation model based on NLLB-200-1.3B, supporting translation between English and 5 African languages

A human-aligned chat model supporting Serbian and English, adapted from Llama-2-7b and trained via Direct Preference Optimization

This is a sentence transformer model based on the LaBSE architecture, specifically fine-tuned for Russian-Mari language pairs, capable of mapping sentences and paragraphs into a 768-dimensional dense vector space.

MalayaLLM is a language model focused on generative AI for Malayalam, built through continued pre-training on the LLAMA2 model with the addition of approximately 18,000 Malayalam vocabulary entries.

NLLB-CLIP-SigLIP is a multilingual vision-language model that combines the text encoder of the NLLB model and the image encoder of the SigLIP model, supporting 201 languages.

OpenLID is a high-coverage, high-performance language identification model supporting 201 languages.

SEA-LION is a series of multilingual large language models optimized for Southeast Asia, supporting 11 Southeast Asian languages, built on the MPT architecture.

Pre-trained multilingual translation model supporting bidirectional translation between English and multiple languages including Kurdish, Samoan, Xhosa, etc.

PhayaThaiBERT is a foundational BERT model based on the Thai language, suitable for Thai text processing tasks.

A general-purpose model supporting over 150 languages, suitable for various natural language processing tasks

A Central Mnong text-to-speech model developed by Meta, supporting conversion of text to natural speech

A Tai Dam text-to-speech model developed by Meta, which is part of the Massive Multilingual Speech (MMS) project.

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

Malayalam text-to-speech model in Facebook's MMS project, implementing end-to-end speech synthesis based on VITS architecture

English-to-Bengali translation model fine-tuned on Helsinki-NLP/opus-mt-en-inc

This model is a fine-tuned Arabic-to-English translation model based on facebook/nllb-200-distilled-600M, achieving a BLEU score of 63.3172 on the evaluation set.

A Norwegian version fine-tuned based on Meta's Llama 2 13b Chat model, optimized for Norwegian text understanding and generation

A 1.3 billion parameter multilingual generation model specifically optimized for Tatar language, deeply fine-tuned based on the mGPT-XL architecture

Nepali speech recognition model fine-tuned based on Facebook's wav2vec2 model