1. A 10 year old Xeon is all you need

    Or running Gemma 4 on a 2016 Xeon with no GPU, 25 flags, 128 GB of DDR3, and a 25B-parameter MoE.

  2. Quantizing Gemma 4's MTP drafters

    Lab notes on getting Gemma 4's MTP drafters converted, quantized, and paired with a verifier.

  3. Open weights are not open source

    On the gap between published model weights and a working local deployment.

  4. Another NixOS Election Cycle

    A cure to resume driven governance?