Troy, Michigan--(Newsfile Corp. - November 4, 2026) - Riemann Computing Inc. has published a technical account detailing the development of its 13.1-billion-parameter language model, My Name is Riemann, providing information about the model's staged growth process, training infrastructure, evaluation results and areas for further development.

Published on October 28 through the OpenPeer AI organization on Hugging Face, the account describes how the model was developed from the published weights of OpenPeerLLM, a 163-million-parameter model. Rather than training a new model from the beginning, the development process expanded the original model through successive stages, reaching approximately 13.1 billion parameters in the current prerelease.

The development process used the OpenPeer-AI-NTK-Trainer, which applies a neural tangent kernel-based gate controller to adjust model outputs without modifying the underlying base weights. According to the published technical account, the approach was used to support experimentation across successive model-development stages.

The publication also documents the computing resources used during development. Initial stages were carried out using a laptop with an external GPU, followed by later stages on a cloud computing node equipped with eight AMD Instinct MI325X GPUs. The company reports that the final model-development stage involved approximately 2.16 billion training tokens.

The technical account includes evaluation results from a held-out dataset of 1,000 prompt-and-completion pairs. It reports a reduction in negative log-likelihood from 1.323 to 1.310 after applying the stage-six controller, alongside a change in token accuracy from 67.73% to 67.84%. The company notes that these results reflect performance on the evaluation data and should not be interpreted as evidence of general-purpose or frontier-level AI capabilities.

The publication also identifies limitations in the current prerelease, including the need for broader evaluation, additional training data and improvements to the training process. The company describes the model as an early development release intended for experimentation and further development rather than a finished consumer chatbot.

The project forms part of Riemann Computing's broader work through OpenPeer AI, which provides publicly accessible AI development projects for researchers and developers. The company's wider technology interests include data compression, computing infrastructure and proprietary enterprise systems.

"Our vision is to create an ecosystem rather than being the sole authority in the AI revolution," said Andrew Magdy Kamal, founder of Riemann Computing.

The My Name is Riemann prerelease and its associated development materials are available through the OpenPeer AI organization on Hugging Face: https://huggingface.co/OpenPeerAI.

Riemann Computing's ongoing business and fundraising information is available at https://wefunder.com/riecomp.

About Riemann Computing Inc.

Riemann Computing Inc. was founded in 2018, initially under the name Stark Drones Corporation. The company works across data compression, computing and decentralized infrastructure, with a strategy that combines publicly accessible AI development projects and planned proprietary enterprise technologies. Learn more at Company Website: https://www.riecomp.org/.

To view the source version of this press release, please visit https://www.newsfilecorp.com/release/322476

Source: ReleasePR.com