2.5x faster inference with Qwen 3.6 27B using MTP - Finally a viable option for local agentic coding - 262k context on 48GB - Fixed chat template - Drop-in OpenAI and Anthropic API endpoints

In my initial post, I mentioned using turboquants. However, I forgot to include instructions for building llama.cpp with the corresponding PR. The PR is currently too unstable and there are animated discussions around it