Nice, is this going to allow a practical use for video games? I know you can do network calls with any model but the latency limits the use cases. If it can run locally it opens up all kinds of use cases like behavior AI for NPCs. Will need to take a look.
.aamt is what GGUF was for the CPU, built for the GPU. And it holds more than model weights.
One format. Weights, memories, worlds. Try it today via npm or github
Nice, is this going to allow a practical use for video games? I know you can do network calls with any model but the latency limits the use cases. If it can run locally it opens up all kinds of use cases like behavior AI for NPCs. Will need to take a look.
.aamt is what GGUF was for the CPU, built for the GPU. And it holds more than model weights. One format. Weights, memories, worlds. Try it today via npm or github