8 docs tagged with "ollama"

Era VIII (M74): The fleet learned the hard way that the right tool and the right hardware are not the same thing. Ollama. ROCm. GTT. Vulkan. The Cogitator speaks its truth.

How using Claude as an AI orchestrator exposed the full depth of AMD ROCm's hardware gaps — and why investing in remote AI intelligence today is the surest path to running sovereign local models tomorrow.

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.

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 V — M56. Maxwell GM204 refuses nouveau. Refuses nvidia-open. The fleet chooses VFIO passthrough and wins. CUDA online. Tzeentch stirs.

Era V — M61. Tzeentch's brain (qwen2.5:14b) is tasked to audit all neurons and recommend model removal. 12 models removed fleet-wide. 100% correct decisions. Zero hallucinations.

The fleet began a model competition with clean assumptions — Ollama, AMD GPU, two test questions. Within hours everything looked broken. Thai text in English responses. 113-second timeouts. A KDE crash. The hardware was fine. The setup was not.

How we tamed gemma4:12b-mlx verbosity from 1537 tokens to 59 — and what it revealed about system prompt engineering.