Riemann Computing Details 13.1B-Parameter Model

The Troy, Michigan company published a technical account of its My Name is Riemann model built on OpenPeerLLM.

Updated on Nov. 4, 2026 in Artificial Intelligence

Photograph of a closed silver laptop beside a matte black external GPU enclosure on a wooden desk near a window, with a blurred server rack behind.
Riemann Computing published a technical account of its 13.1-billion-parameter My Name is Riemann model, built by expanding a 163-million-parameter base through successive training stages. AI Illustration. Upload story photo >

Riemann Computing published a technical account of its 13.1-billion-parameter My Name is Riemann model on October 28 through the OpenPeer AI organization on Hugging Face. The company describes it as an early development release intended for experimentation.

Why it matters

The account documents how a much larger model was built by expanding a 163-million-parameter base through successive training stages rather than starting from scratch. Readers following open model development can see the specific hardware, data and evaluation figures behind the release.

The model grew from OpenPeerLLM's 163 million parameters to about 13.1 billion, trained on roughly 2.16 billion tokens in the final stage using eight AMD Instinct MI325X GPUs. Evaluation on 1,000 held-out prompt-and-completion pairs showed a small drop in negative log-likelihood from 1.323 to 1.310, and the company has not reported results on public benchmarks.

The players

Riemann Computing

The Troy, Michigan company that developed the My Name is Riemann model and published its technical account.

OpenPeer AI

The organization on Hugging Face that hosts the published weights and development materials for the OpenPeerLLM family of models.

The details

Development used the OpenPeer-AI-NTK-Trainer, which applies a neural tangent kernel-based gate controller that adjusts outputs without modifying the base weights. Work began on a laptop with an external GPU before moving to a cloud computing node for later stages.

Timeline

  1. October 28, 2026: Technical account published on the OpenPeer AI Hugging Face organization.

  2. November 4, 2026: Press release announcing the technical account distributed.

The Tech Race

The My Name is Riemann model extends the publicly released OpenPeerLLM weights, which were 163 million parameters, into a 13.1-billion-parameter system through staged expansion rather than training from scratch. The approach shows how smaller open base models can be grown in size by a single developer team using a stage-by-stage controller.

Developers who experiment with open models can review the published training stages, hardware setup and evaluation figures to judge whether the approach fits their own work. Because the model is an early prerelease, its practical use for everyday applications remains limited until further results are published.

The takeaway

Readers interested in open model development can use the published stage-by-stage figures as a reference point for judging smaller-scale expansion efforts. The release is labeled an early development version, so claims about its capabilities should be weighed accordingly.

Further reading

For related coverage, see Artificial Intelligence on the Troy technology news page.

More information

Review the OpenPeer AI Hugging Face model and development materials for the published weights and technical details.