Spaces:

---

ethicalabs

/

Echo-DSRN-114M-Base

like 0

Sleeping

I have stabilized the Echo-DSRN (Dual State Recurrent Neural Network) at the 114M parameter scale.

It uses a surprise-gated continuous memory state that mathematically compresses long-context history into a strict O(1) memory footprint.

I am requesting a GPU grant to scale this proven architecture out of the 114M "wind tunnel" phase and pre-train the ~0.5B and ~1B parameter classes.

The goal is to provide the open-source community with a sovereign, edge-capable routing model that can process large context without crashing consumer hardware.

The baseline architecture is available at: https://huggingface.co/ethicalabs/Echo-DSRN-114M-v0.1.2-Base