Andrea Borio

I play with LLMs.

Inference architecture

All field notes

Sparse attention, transactional state, fail-closed admission, and storage at checkpoint scale—short decision records with evidence and explicit limits.

  1. Sparse attention · AFN-016 A sparse mask is not sparse attention
  2. Transactional state · AFN-010 One bank flip is enough: make the whole inference chunk transactional
  3. Storage · AFN-003 320 million embedding rows without a 320 million-entry index
  4. Admission · AFN-004 Admission before Metal: make every bad artifact fail cheaply

Recent posts

All posts
  1. Why this blog exists

    The useful part of an experiment usually happens after the screenshot.

  2. I tried to run GLM-5.2 on a 64GB Mac

    Field notes from an experimental ds4 fork, a 244GB GGUF, and the small horror of sparse models that are sparse in compute but not very friendly to filesystems.

  3. Re-quantizing a local model, 14× faster

    Where a 2-bit model spends its bits, and why trying answers used to cost eighty minutes