- [x] `llama-server` - [x] `llama-server router` - [x] gpu only works - [x] gpu+cpu offload works (copy any existing quant from titan or lithium) - [x] `gemma4 12b qat mtp` - [x] `gemma4 12b qat mtp` all modalities - [x] qwen3.6 35B moe - [x] qwen3.6 27b - [x] gemma4 26b - [x] gemma4 31b All 10 bullets resolved. See [milestone-PASS comment](https://github.com/heiervang-technologies/ht-llama.cpp/issues/100#issuecomment-4661301507) for per-bullet evidence + bench JSON corpus on PR #99.
llama-serverllama-server router(copy any existing quant from titan or lithium)
gemma4 12b qat mtpgemma4 12b qat mtpall modalitiesAll 10 bullets resolved. See milestone-PASS comment for per-bullet evidence + bench JSON corpus on PR #99.