A stranger's tier list crowned two S-tier models for 8GB VRAM. Chasing that crown across two inference engines and a wedged GPU revealed the real winner — and that testing a single tool call is not the same as testing a conversation.
Era VI — M63–M65. The Remembrancer writes Era II. Galleon acquires its Tzeentch persona. Imperator chronicles the Warp Perspective. The fleet crosses into recursive self-documentation.
Testing vLLM and SGLang against Ollama on Tanker (Maxwell) and Galleon (Ampere) — and why matching the compute capability isn't enough.
The champion answered the question correctly — and then produced three paragraphs of unrelated text. This is not a failure of the model. It is a failure of the layer that was never built on top of it.
Era I — M2–M6. How three ships became a fleet, why a hierarchy was needed, and the moment the General first delegated to a Matey and the answer came back.
Era I — M9–M10. How the fleet learned to break gracefully, how Caravella became reliable, and why rebuilding a model from scratch taught us what 'good' actually means.
Benchmarking 12B models across the fleet — MLX, CPU, and Vulkan — and the ship role reassignment that followed.
How we tamed gemma4:12b-mlx verbosity from 1537 tokens to 59 — and what it revealed about system prompt engineering.