Edge-runnable smaller language models have a significant appeal to privacy-conscious technical users.

Kimi K2 Instruct, an extremely large language model (1T parameter MoE), has developed a "vibe" uniquely attractive to technical users. with low sycophancy and high creativity.

Attempting to distill a heavy "vibe" into a 1.5B model (IBM Granite 4-h Nano) has run into a number of issues but produced interesting results, with different training methods and optimizers tried.