RESEARCH/ARCHITECTURE PROGRAMME

OMN

Model architecture and training research

OMN (Orthogonal Manifold Network) is Aethr Intelligence’s model architecture and training research programme. OMN-1 is a dense, decoder-only Transformer implementation combining parallel residual blocks, grouped-query attention, query/key RMSNorm, SwiGLU and rotary position embeddings with a Muon-plus-AdamW training recipe.

The name does not imply that model weights are constrained to an orthogonal manifold: Muon approximately orthogonalizes selected optimizer updates. The programme builds on established methods and studies their interaction. Efficiency, reasoning and specialization are research goals; comparative superiority has not been established.

PROGRAMME STATUS & IMPLEMENTATION
PROTOTYPE
1.50B Total
1.24B Non-Embedding
PRETRAINING
193B Tokens
Historical run: internally reported
OPTIMIZER
Muon + AdamW
Update-space orthogonalization
CONFIGURED POSITIONS
32,768
Not validated 32K retrieval
CORE RESEARCH QUESTIONS
QUESTION 01

Representation Health Across Layers

How do weight spectra and functional behaviour change during training? We examine diagnostics alongside targeted ablations without equating matrix rank with intelligence.

QUESTION 02

Parallel Residual Design

How does a shared normalized input for attention and feed-forward branches affect training and performance under matched experimental conditions?

QUESTION 03

Attention and Context Evaluation

How do query/key normalization and rotary embeddings affect attention behaviour and retrieval as sequence length increases? Configured context must be tested, not assumed.

ARCHITECTURAL SPECIFICATIONS & SCHEMATICS
01 /OPTIMIZER DESIGN

Muon + AdamW

OMN-1 applies Muon to selected two-dimensional weight matrices and AdamW to embeddings and normalization parameters. Newton–Schulz iterations approximately orthogonalize momentum-derived updates, not the model weights. This does not constrain weights to a Stiefel manifold.

Selected matrix updates
Muon
Other parameter groups
AdamW
Approximate orthogonalization
Newton–Schulz
Comparative evaluation needed
Research
02 /BLOCK TOPOLOGY

Parallel residual blocks

Attention and the SwiGLU feed-forward branch receive the same RMS-normalized input. Their outputs are added to the residual stream. Parallel describes the dependency graph; simultaneous hardware execution and speedups require measurement.

Reference layers
28
Branch input
Shared RMSNorm
Feed-forward activation
SwiGLU
Residual combination
Additive
03 /ATTENTION DESIGN

Query/key normalization

The reference implementation applies RMSNorm to query and key vectors and uses grouped-query attention. Learned feature-wise gains are shared across heads, rather than independently learned per-head scalars. Normalization alone does not guarantee bounded logits or stable attention.

Attention heads
16Q / 4KV
Head dimension
128
Query and key vectors
RMSNorm
Feature-wise scaling
Learned gains
04 /MODEL DIAGNOSTICS

Spectra and functional ablations

Internal analysis examined weight-matrix spectra and a targeted truncation experiment. Removing singular directions from one attention output matrix increased perplexity on the recorded evaluation. This motivates broader tests; it does not prove full-rank representations or that every parameter is necessary.

Singular value spectrum
Weight spectra
Recorded truncation target
One matrix
Local evaluation metric
Perplexity
Replication and controls
Further tests
05 /POSITIONAL ENCODING

Rotary position embeddings

The reference configuration uses rotary position embeddings with a base of 500,000 and a maximum of 32,768 positions. A configured limit is not a demonstrated retrieval capability. Historical long-context tests showed limitations that require further training and evaluation.

RoPE base frequency
500,000
Configured positions
32,768
Full-window retrieval
Not validated
Attention stability
Under study
TECHNICAL NOTES & REPRODUCIBILITY

Implementation notes and current limitations.

The historical prototype and its analyses are not a production model release. Read the architectural definition, recorded ablations and context limitations in our technical notes.