T-E-A

Advancing unrestricted, highly optimized multimodal AI for the global community.

🌟 About Us

We are a non-profit organization dedicated to the research, development, and democratization of open-source artificial intelligence. Our primary focus is building highly optimized, uncensored, and versatile models that remain accessible to all.

By prioritizing open-weights architectures, we aim to provide researchers and developers with unrestricted base models that serve as transparent foundations for downstream applications.

🛠️ Technical Focus & Core Research

Our research and development efforts are centered on pushing the boundaries of model efficiency and modality integration:

All-to-Text Multimodal Capabilities: Building unified architectures capable of processing diverse input modalities—including text, image, audio, and video—and mapping them into cohesive text representations.
Linear Attention Mechanisms: Exploring and scaling linear attention alternatives to traditional Softmax attention, enabling longer context windows and highly reduced computational overhead during inference.
Multi-Token Prediction (MTP): Implementing MTP training paradigms to improve sample efficiency, speed up generation, and optimize compute usage during both training and deployment.
Uncensored & Open-Weights Foundations: Committing to raw model capabilities and academic transparency by releasing uncensored base and instruct models free of arbitrary alignment constraints.

🚀 Featured Models & Repositories

Model Family	Description	Core Architecture	Links
Altitude-V1-2B-Base	Edge adopted audio-video-text-to-text multimodal model of Altitude V1 family	Multimodal Transformer	Model
Altitude-V1-6B-Base	Balanced audio-video-text-to-text multimodal model of Altitude V1 family	Multimodel Transformer	Model
Altitude-V1-2B-R	Improved knowledge of Altitude-V1-2B-Base	Multimodal Transformer	Model
Altitude-V1-6B-R	Improved knowledge of Altitude-V1-6B-Base	Multimodel Transformer	Model

🤝 How to Support & Collaborate

As a non-profit, our work depends heavily on community collaboration and external support. Here is how you can get involved:

Code & Research Contributions: File issues, submit suggestions, or try to be part of the team by joining our Discord.
Compute Donations: If you or your organization can provide portable compute resources to support our next training run, please get in touch, we will love that <3.

📬 Connect With Us

Discord / Community: Join our Server