PowerDevHouse

We're excited to introduce @Gratis_AI. Run opensource AI models locally on your machine for free, with no setup required. Gratis runs models from @huggingface directly in a browser tab or desktop app for more power, making every chat private by design. TPN builds small, capable models; gratis runs them. gratis-ai.com

Upcoming: TPN Product Launch In November, we’ll launch a consumer facing product powered by TPN. The product will serve as one of the distribution layers for TPN Labs models, acting as an interface between our models and every day users. We are excited for what's to come!

Next month is going to be a big one for TPN 👀 We’ve been building something that we want to share with the world. Powered by TPN_Labs - Subnet 65 $TAO

We asked for a model that fits on a laptop and knows things. A case study in purpose-built model optimization.

Here’s the case study for the TPN-005 model. We asked the network for a purpose built LLM that could fit inside a chat interface’s memory and perform well on general knowledge. The subnet produced a successful output, a model optimized for general knowledge within a fixed memory budget, performing well on the requested capability. Every figure in the study links to an immutable run record, from the model revision, sealed epoch, to the evaluation settings, so everything is verifiable.

TPN Labs produced its first release-ready model. The model (TPN005) is the outcome of an optimization process that reduced the size of the Ministral-3-8B base model by ~40%, while keeping its full 32k context. This was the first round scored entirely on Fluid Bench, and it marks the first successful TPN upgrade. So, what now? We are happy with this result. This was Fluid Bench's first iteration, which paved the way for even better models to be produced by Subnet 65, with many more patches and upgrades to be implemented along the way. Subnet produced models will become highly relevant once we release TPN's upcoming consumer facing product. To close Competition 5 on a high note, we'll release a case study with all the numbers shortly.

TPN's new benchmark: fluid_knowledge Benchmarks like MMLU, HellaSwag, and ARC have been public for years, which means they're in the training data of nearly everything. A high scoring model could be one that understands the material, or a model that memorized the answers to the test. The test result looks the same, but the performance between the two varies significantly. This is why we built fluid_knowledge, a benchmark whose questions we generate ourselves and never publish, so that models can't train for it in advance. Available to anyone now on TPN Bench.