https://talachu.com/blog/ Notes on small models in production — memory math, economics, and new releases. en Fri, 30 Oct 2026 00:00:00 GMT https://talachu.com/blog/#embeddinggemma-2 https://talachu.com/blog/#embeddinggemma-2 Tue, 06 Oct 2026 00:00:00 GMT 740M, Apache 2.0. Text-only use fits in about 191 MB at 4-bit on a Pixel 11 Pro. https://talachu.com/blog/#one-model-nine-quants https://talachu.com/blog/#one-model-nine-quants Fri, 02 Oct 2026 00:00:00 GMT The same model at FP16, Q8_0, Q6_K, Q5_K_M, Q4_K_M, Q3_K_M, Q2_K — and what each one costs in memory.