b.BEKHRUZ SULEYMAN

01 / Research archive

Research

Language models, attention and gating mechanisms, and experiments that can be reproduced under constrained compute.

Active research

CURRENT DIRECTION
R.01active

Alethic

Language-model architecture research, from controlled small-model experiments to a planned ~1B Phase I.

~1B target
R.02experimental

SGCA

Sin-Gated Control Attention: separating attention retrieval from learned residual control.

Architecture research

Completed experiments

PRELIMINARY EVIDENCE
R.03completed

SinGatedLM

Parameter-matched small language-model experiments with a sinusoidal attention gate.

~64K / ~1M

Models & releases

Model index

Alethic-151M

Experimental pretraining · SinGatedAttention lineage

151M / Experimental

116M predecessor

Historical base model · Released under the Ruzz name

116M / Historical

Research writing

All writing

Why separate retrieval from control?

Research · sgca · research
2 min read

What the small SinGated experiments actually show

Experiments · experiments · research
2 min read

Planning a 1B language-model experiment independently

Training · alethic · training
2 min read

Papers & reports

The research pages document the mechanism, methods, and reported evidence. No DOI or peer-reviewed publication is claimed. Complete Phase-I artifacts will be linked when available.