Alignment Lab : Augmented Generative Inference
AI & Machine Learning
< status : funded >
Project Details
Continued research and development into a series of efforts which have produced a measurable gain in the heightened quality of Local LLMs, and development towards a more accessible ecosystem for user development and curation of their own personal models.
To allow for a fully automated user curated cycle of training and dataset generation on affordable local hardware of a model which can hear, see, and read the users inputs locally and respond in kind with the same level of understanding and expressiveness that the user can.
To allow for a fully automated user curated cycle of training and dataset generation on affordable local hardware of a model which can hear, see, and read the users inputs locally and respond in kind with the same level of understanding and expressiveness that the user can.
Objective(s)
To allow for a fully automated user curated cycle of training and dataset generation on affordable local hardware of a model which can hear, see, and read the users inputs locally and respond in kind with the same level of understanding and expressiveness that the user can.
Technical Specification (including tools and technologies to be used):
Currently continuing research started with Bakllava, to create mass scale labeled datasets containing over 90 multimodal sentiment labels generated with bimodal clip variants to place the labels orthogonally in latent space, and in temporal relationship to the emotive events occurring in the data pipeline, additionally pursuing research which appears to allow for the generation of images and audio from embeddings derived from decoder only models like mistral.
Budget Breakdown
Whatever it takes, currently able to boast an extremely efficient self developed pipeline able to do large scale data operations for free, and a custom trainer which can perform experiments, fine tunings, and peft methods at roughly half the wall clock time and price.
Project Timeline
Completion 80% timeline ~40 days
Team Background
Alignment Lab AI is a distributed team of developers and engineers from all over the planet, notably each founder has a following and is popular in their own right despite the lack of explicit connection to alignment lab on the surface.
Project Impact
Multimodality is oncoming, this project may be the first or one of the first but the margins wont be giant, the opportunity that this project presents for the decentralization cause is that it instantiates a standard for this type of communication to be done locally, and not over an API. Data provenance is important. Keep it, store it, dont share it, package it, clean it, and sell it. If were lucky enough people recognize the value of their data that the market demand takes care of us and both utility and money can be generated in a private way, where human generated data is only given consensually, and compensated fairly.
Alignment Lab
For more info and inquiries, please visit the links below
Building a project and looking for funding?
Grant Sponspor & Partners
Experience the future of Telegram community engagement.