OpenELM is a set of open-source efficient language models that employ a hierarchical scaling strategy to optimize parameter allocation and improve model accuracy. It includes four parameter scales: 270M, 450M, 1.1B, and 3B, offering both pre-trained and instruction-tuned versions.

Natural Language ProcessingTransformers

MADLAD-400-3B-CT2 is a multilingual machine translation model based on the T5 architecture, trained on public data of 1 trillion tokens covering over 450 languages. This model supports over 400 languages, focuses on machine translation and multilingual NLP tasks, and is also competitive compared to larger-scale models.

Natural Language ProcessingTransformersMultiple Languages

MADLAD-400-7B-MT is a multilingual machine translation model based on the T5 architecture, trained on 250 billion tokens covering over 450 languages. The model has 7.2 billion parameters, supports machine translation for over 400 languages and multilingual NLP tasks, and its performance is comparable to significantly larger models.

Natural Language ProcessingTransformersMultiple Languages

BioCLIP is a vision foundation model based on the Tree of Life, adopting the CLIP architecture as a general-purpose visual model for biology, covering over 450,000 taxonomic units.

MultimodalOpenclipEnglish

GENA-LM is an open-source foundation model for long DNA sequence analysis. It is a masked language model based on the Transformer architecture, specifically designed for the field of bioinformatics and capable of processing DNA sequences up to 4500 nucleotides long.

Scientific ComputingTransformersOther

A large-scale English automatic speech recognition model based on the Wav2Vec2 architecture, fine-tuned on 4500 hours of multi-source speech data, supporting decoding with a language model

Audio ProcessingTransformersEnglish