Yes, this is real AI. These are actual Transformer models, not just scripted responses or fake AI. This is the smallest PocketLM model that can actually produce coherent text. For reference, the smallest model I found that speaks reliably is around 1 billion parameters (100,000 times bigger). It’s extremely fast—so fast that it doesn’t even need TurboWarp! Despite its tiny size, this model is actually usable and surprisingly coherent. It’s small, but it’s not completely dumb. It burns through context ridiculously fast, lol.
Let's get this on trending! Add to all the studios! It's an exponential curve!!! :))) I performed a parameter/configuration sweep to find the best-performing configuration for PocketLM 10K. This model is part of the PocketLM family (testing branch): PocketLM on GitHub: https://github.com/Greninja9257/PocketLM Huge thanks to everyone working on and experimenting with the PocketLM project! (mostly me, tho)
