About our Research

An end-to-end, mission-focused approach to GenAI

We take an end-to-end, mission-focused approach to GenAI research, training foundation models, enhancing the technical possibilities with generative AI and developing systems that reflect the values we work to ingrain.

We are committed to responsible and open research. We raise the bar on openness, responsibly sharing all innovations without obfuscation.

Highlight Contributions

Research

Introducing Pharia-1-LLM: transparent and compliant

We are pleased to announce our new foundation model family that includes Pharia-1-LLM-7B-control and Pharia-1-LLM-7B-control-aligned, now publicly available under the Open Aleph License, which explicitly allows for non-commercial research and educational use.
Read more
Research

In awe at the scale of these tensors – a gentle introduction to Unit-Scaled Maximal Update Parametrization

Together with Graphcore, we recently developed u-μP as a new paradigm to parametrize neural networks in terms of width and depth. Our approach combines μP, developed by G. Yang et. al., with Unit Scaling, a concept introduced by Graphcore.
Read more
Research

T-Free: Hierarchical Autoregressive Transformers for Language Fairness and Sovereignty

In this blog post, we want to take a closer look at a tokenizer-free approach, which we proposed in a recent paper and termed Hierarchical Autoregressive Transformers (HAT). In particular, we want to showcase how such a model can be pre-trained in English and efficiently adapted to learn a new, previously unseen language.
Read more

Our Principles

While our research is foundational, it always aims to address real industry needs & set standards for GenAI applications.

We are responsibly sharing our innovation, often by publishing them including papers, blog posts & codebases. This allows our research to be transparent and visible to the community.

Teams are mostly self-organizing and consist entirely of researchers and engineers. This way we ensure all voices are heard and are able to maximize technical influence of individual contributors, regardless of career level.

AlphaOne, our datacenter, is one of the most powerful in Europe, and is managed by the Aleph Alpha infrastructure team. This way, we are able to to provide our team a high GPU/head ratio and experimental/ innovative hardware.

Research Contributions

Research

T-Free: Hierarchical Autoregressive Transformers for Language Fairness and Sovereignty

In this blog post, we want to take a closer look at a tokenizer-free approach, which we proposed in a recent paper and termed Hierarchical Autoregressive Transformers (HAT). In particular, we want to showcase how such a model can be pre-trained in English and efficiently adapted to learn a new, previously unseen language.
Read more
Research

In awe at the scale of these tensors – a gentle introduction to Unit-Scaled Maximal Update Parametrization

Together with Graphcore, we recently developed u-μP as a new paradigm to parametrize neural networks in terms of width and depth. Our approach combines μP, developed by G. Yang et. al., with Unit Scaling, a concept introduced by Graphcore.
Read more
Research

Words don’t come easy (… to LLMs): Universal Text-Encoding for dynamic, multi-lingual alphabets revolutionizing efficiency and effectiveness for LLM training and inference

The remarkable advancements of Large Language Models (LLMs) frequently capture attention as they become valuable collaborators in daily situations, all while progressing towards breakthroughs beyond simple language completion.
Read more
Research

Introducing Pharia-1-LLM: transparent and compliant

We are pleased to announce our new foundation model family that includes Pharia-1-LLM-7B-control and Pharia-1-LLM-7B-control-aligned, now publicly available under the Open Aleph License, which explicitly allows for non-commercial research and educational use.
Read more
Research

Open-sourcing Codebase Scaling for Non-commercial Research

Aleph Alpha’s model training codebase Scaling is publicly available under the Open Aleph License, which explicitly allows for non-commercial research and educational use. Scaling was used to develop our concurrently released new models Pharia-1-LLM-control and Pharia-1-LLM-control-aligned.
Read more
Research

Quality Diversity through AI Feedback

Language models carry implicit distributional biases based on their training data, which can reinforce existing norms. In this work, we take one step towards addressing the challenge of unwanted biases by enabling language models to return outputs with a broader spectrum of attribute traits, specified by a user. This is achieved by asking language models to evaluate and modify their outputs.
Read more
Research

Luminous Performance Benchmarks

The research compares Luminous to the models from GPT-3 and ChatGPT developer OpenAI, among others. The scientific comparison included tasks related to text classification, evaluation, and generation, as well as answering questions about specific text contents. The result is impressive – with Luminous, a European AI language model is, for the first time, on par with the world’s leading AI language models, while being much more efficient.
Read more
Research

Aleph Alpha and Graphcore demonstrate 80% sparsified AI Model

Aleph Alpha and our partner Graphcore are unveiling a significant advance in AI compute efficiency, with the sparsification of a 13bn parameter model down to just 2.6bn parameters.
Read more
Research

Luminous-Explore – A model for world-class semantic representation

AI becomes meaningfully more capable almost every month. With this new fidelity and transformative use-cases, society is rethinking the human-machine-collaboration and ethical alignment.
Read more

Join the Aleph Alpha Team

Be part of a wild, ambitious and high performing team, of people living their best work life 

Our Tenets

We own work end-to-end

We take responsibility for every stage of the process, ensuring that our work is complete, scalable, and of the highest quality. We embrace accountability and continuously seek improvements along the way.

We ship what matters

Our focus is on solving real, immediate problems for our customers and the research community. We prioritize delivering impactful solutions that bring value and make a difference.

We work transparently

We share our work openly to our partners, customers, and the broader community through publishing and sharing results and insight including blogposts, papers, checkpoints, and more.

We innovate through our passion and talents

We strive to balance ideas and interests of our team with our mission-backwards approach, and acknowledge the interdisciplinary, diverse perspectives in our team.