Tag
GuppyLM is a small ~9M-parameter language model that can be trained from scratch in 5 minutes on a single Colab notebook. It simulates a fish named Guppy talking about water, food, and light, aiming to prove that training your own language model isn't magic — it's something anyone can do.
qev-0.8b is a 0.8B-parameter local decision model that takes a single forward pass to return calibrated probabilities over any input, ranking #1 of 27 similar-sized models on Benchmark Heaven (77.1%) at 3.5 ms p50 latency while running in 1.3 GB on a laptop. It ships in CUDA, MLX and GGUF formats and generalizes to games (Tetris, Pac-Man, 2048, Breakout, Whac-A-Mole) never seen in training.
A developer demonstrates a 24KB stand-alone HTML-LLM that can generate consistent stories at over 60 tokens per second on a smartphone, achieved via adaptive quantization and QAT on the 81KB MacroStory model using local Qwen3.8 tooling.
PrismML showcases its tiny 1-bit Bonsai LLM at Qualcomm's Snapdragon Summit, which can run locally on smart glasses powered by the Snapdragon AR1 Gen 1 Platform, enabling real-time vision and language processing.
PrismML, a Caltech-founded startup, released Bonsai 2, a highly compressed LLM that reduces memory usage by 9-10x with minimal performance loss, aiming to enable AI on consumer devices like PCs and smartphones.
A tiny ~10K parameter router called tinyrouter learns which open model to use per question on MMLU, outperforming individual models by optimizing allocation.