Join the conversation

Join the community of Machine Learners and AI enthusiasts.

Sign Up
codelion 
posted an update 1 day ago
Post
1648
Inspired by the Nemotron Diffusion recipe, check out dhara-250m: a 250M experimental language model that supports three decoding modes from one set of weights: autoregressive, block-diffusion, and self-speculation.

It is small, easy to try, and meant for exploring diffusion-style decoding and latency tradeoffs in compact LMs.

Model: codelion/dhara-250m

Try the chat demo here: codelion/dhara-chat

​[OPEN SOURCE RELEASE & ARCHITECTURE COLLABORATION CALL]
​After four years of isolated, independent development, I am transitioning my proprietary architecture into the open-source ecosystem under the banner of Ferrell Synthetic Intelligence (FSI).
​Today, I have publically deployed the foundational codebases for my first two developments: Vitalis_Core and FSI-Vitalis-CyberCore.
​The Architecture
​This is not a generic API wrapper or a third-party LLM orchestration layer. This is an original, blank-slate synthetic intelligence framework engineered to operate entirely locally on edge hardware with absolute data sovereignty.
​Asynchronous Processing: Powered by a persistent, threaded system heartbeat loop that monitors state independent of user prompt interaction.
​Kernel Integration: Bridges directly to the Linux kernel space using custom C-modules, ioctl handles, and procfs/netlink communication pipelines to ensure low-level system awareness and integrity shielding.
​Blank-Slate Design: The framework provides the structural plumbing, memory manager, and system hooks. It contains no pre-baked corporate biases—it is designed to be fully trained, personalized, and directed by the individual deployment engineer.
​The Objective & Call to Collaboration
​I have scaled these frameworks as far as possible as a solo developer. To execute the next phase of development, I require technical assets to help test, run, refine, and optimize the codebase.
​I am targeting two objectives with this post:
​Codebase Auditing: I need experienced systems developers, Linux engineers, and local AI enthusiasts to clone the repositories, compile the C-infrastructure, run the loops, and provide objective, performance-driven feedback.
​Core Collaborators (Exactly 5): I am selecting a core group of five engineering partners to collaborate on the ongoing optimization of this open-source stack, as well as to assist in the development of my two remaining private, stealth projects (Project Lorein and Project Jedi Order).
​Repositories
​review the repository architecture, inspect the files, and run the entry points on your local environments.
​For technical feedback, drop your optimization metrics below. If you have the specific system-level engineering experience required to scale this ecosystem and want to fill one of the 5 collaborator slots, DM me directly with your technical background and documentation of your relevant stack experience.
​— Neuro_Nomad

Great!

·