In-House SOTA Foundation Models

NuMind trains task-specific and domain-agnostic foundation models for information extraction that rival frontier models.

Our Open-Source Foundation Models

You can use them out-of-the-box or customize them with NuMind for better performance! Open-sourced under MIT license.

Entity Recognition

Token embeddings for NER in English

NuNER_multilingual-v0.1

Multilingual token embeddings for NER

Classification

Embeddings for sentiment analysis

Embeddings for content moderation

Embeddings for topic classification

Structured Extraction

3.8B parameters

NuExtract-1.5-smol

1.7B parameters

NuExtract-1.5-tiny

0.5B parameters

Zero-shot Entity Recognition

Zero-shot token classifier

NuNER_Zero-span

Span-prediction version of NuNER Zero

NuExtract_Zero-4k

Long-context version of NuNER Zero

Join Our Discord Check on Hugging Face

Our Private Foundation Models

Only available via NuMind.

NuExtract-2

Structured Extraction with summarization feature

NuHealth

Specialized for healthcare and medical coding

NuCode

specialized for code documentation and other code-related tasks

NuMarkdown

replaces OCR and ensures for an efficient rag

You face a unique NLP challenge? Talk to us to access our un-released foundation models, and get in-house model customization by our model experts.