Skip to main content
EMEP is a documentation-first platform for engineering new large language models from existing compatible open-weight models. The primary loop is deterministic and auditable: analyze compatibility, execute tensor-level merges, evaluate candidates against isolated benchmark splits, compute fitness, and evolve the population. Fine-tuning and RAG are optional supporting layers, not the core.
This site is the specification. Implementation has not started. Every page describes intended behavior with explicit evidence tags: Established / Hypothesis / Assumption / Result / Goal / Unsupported.

​
The Primary Research Loop

​
Explore the Specification

Start Here

Understand EMEP in five minutes and the first workflow you can trace end to end.

Project Constitution

The non-negotiable scientific and engineering rules that govern every decision.

System Architecture

Three-level architecture, from executive view to internal components.

Merge Engine

Tensor-level merging with SLERP, TIES, DARE, Task Arithmetic, and structural merges.

Evolution Engine

Population-based search with single- and multi-objective optimization.

Evaluation Framework

Optimization, validation, and hidden test splits with strict isolation.

​
What EMEP Is Not

Not a chatbot

Not a RAG platform

Not a fine-tuning platform

Not an agent framework

Read the Vision & Scope for the full boundary, and the Glossary for exact terminology used across every specification.