Wikitune is a data curation tool for bridging the living knowledge of wiki's with Decentralized AI training & tuning
Wikitune is a data curation tool for bridging the living knowledge of wiki's with Decentralized AI training & tuning.
Wikitune is an application built to allow Gaianet Node operators of bespoke LLM's to build finetuning sets for their nodes from Wiki's around the web. Whether sourcing from internet stalwart Wikipedia or any other varieties of Wiki's, this application seeks to build tunable sets of data for your specialized LLM usecases.
This application is meant as a support app for the deployment of custom LLM's. It utilizes Decentralized AI (DeAI) protocol, Gaia Network for creation of training set. In the interest of supporting an Living Knowledge ecosystem, we seek to connect one of the original crowd-sourced knowledge evolution machines: The Wiki with decentralized technologies. To us, this seems like not only a natural fit, but an inevitable one.
Although a relatively simple usecase, by addressing the usability of tuning on open source LLM's, Wikitune can be used by communities to build out custom LLM's that serve their bespoke needs.
Moving forward:
This tool was derived from a toolchain we are already looking to expand for building out LLM's (as of now, specifically on the Gaia ecosystem). Moving forward, we will begin investigating possible expansions to allow LLM operators to build amazing bespoke opportunities.
Q4 2024 - Train & Deploy a Gaia Node
In order to scale out our service and increase speed and efficiency, we will be investigating how to utilize DeAI tools to create an infrastructure fabric for high throughput training. We will also seek to acquire a Wikitune Gaia domain and build out a node specific for creating knowledge feedbacks and finetuning.
Q4 2024 - Workbench Feature Expansion
For this early release, the Project workbench has extremely limited functionality. Moving forward, we seek to add set evaluations and improving tuning Q/A quality with fully customizable parameters going into the LLM set.
2025 - Wikis, Wikis, Wikis
This aspect will begin to expand almost immediately but by 2025, the goal will be to service both Open Wiki and Fandom-based wikis so community sourced knowledge can be used for tuning almost immediately.
TBA - DePIN Deployment & Integration
One of the most exciting areas of DeAI (and possibly modern Web3 in general) is the Decentralized Physical Infrastructure Networking. Currently, we are currently investigating building a comprehensive deploy-to-use-to-earn pipeline for building out a system that will enrich the data acquisition, data processing, training, and inference delivery of DePIN opportunities.
First, we began thinking about where we could implement LLM's into our existing business set and had a few internal brainstorm sessions, diagnosing challenges and opportunities alike. For us, Gaia presented an opportunity to help disrupt the consolidation of LLM's. The project's Web3 roots presented both the technological and philosophical base for converging both sectors. Finally, and to be completely honest, I just started to build-- looking towards the past (Wikis) to match the layers that hit the right notes with what we wanted to see and what would help us utilize Gaia to its fullest.
No Fundraising but putting this here because there is only room for one github link here. Clientside codebase can be found here: https://github.com/tai-bcw/wikitune-client