About
What this blog does
Once a month I take a recent paper on serving, observability or efficiency of AI models and try it for real:
- Reproduce the paper’s main claim with the authors’ released code.
- Extend it with one question the paper leaves open.
- Report what held up, what didn’t, and what it cost.
Results are reported as measured, including failures. When I speculate, I say so.
The lab
- Raspberry Pis of several generations, for edge and thermal experiments
- Intel NUCs (i7 and i5) running Proxmox, for CPU inference and the monitoring stack
- GMKtec EVO-X2 with 128 GB of unified memory, for large models on AMD hardware
- Google Cloud GPUs and TPUs, used briefly and with hard cost limits
The full setup is described in the first post.
Code and data
Everything needed to rerun an experiment is in the lab repository, one folder per post.