Skip to content

Latest commit

History

History
50 lines (36 loc) 路 1.13 KB

File metadata and controls

50 lines (36 loc) 路 1.13 KB
title AETHORIA-AI
emoji 馃З
colorFrom blue
colorTo purple
sdk static
pinned false

AETHORIA-AI

Complexity Framework is a PyTorch research stack for deterministic Token-Routed Mixture-of-Experts language models.

Main architecture families

  • TR-GQA: grouped-query attention + TR-MoE.
  • TR-MHA: multi-head attention + TR-MoE.
  • TR-MoE: a dense shared SwiGLU path plus deterministic token-selected experts.

Token identity selects expert parameters. The selected experts still transform the current contextual hidden state.

Research standards

Projects distinguish implementation, pilots, controlled comparisons, and planned runs. Model cards should report:

  • total and active parameters;
  • tokenizer and context length;
  • data and token budget;
  • seeds;
  • evaluation split;
  • hardware and runtime;
  • license and intended use;
  • known limitations.

Repository

https://github.com/Complexity-ML/complexity-framework

Hugging Face organization

https://huggingface.co/AETHORIA-AI

License

The framework repository uses CC BY-NC 4.0. Individual model repositories must state their own license explicitly.