Optimal Settings for Qwen3.6-27B Coding
starkruzr · reddit · 2026-07-18
The poster is discussing the most effective combinations of harnesses, server parameters, and system prompts when using Qwen3.6-27B for coding tasks.
They first shared their llama-server presets used on a machine with 2×16GB GPUs, detailing various pairings of quantization, MTP, KV cache, and context lengths. They summarized a few key findings:
- q80 KV provides roughly 1.5× more context than f16 KV with minimal quality loss.
- MTP (draft spec-decode) speeds up generation but sacrifices some context length.
- A 32GB VRAM setup can essentially only load a single 27B model at a time, and switching models incurs a few seconds of reload overhead.
The core question is: given these hardware constraints, how are others configuring their harnesses, system prompts, and inference settings to make this model act as a reliable coder?
More from coding & agent
- Ben Bajarin: Frontier AI is becoming the enterprise cyber-defense stack — BenBajarin · 2026-10-26
- Building AI bots on raw APIs? Devs say models are 'dumb without the harness' — madeofAjala · 2026-10-26
- HarnessMM cuts agent tokens per task by up to 88% while raising success rates — omarsar0 · 2026-10-26
- The CLI is no longer the ideal interface for AI agents, argues AI communicator Elvis Omarsar — omarsar0 · 2026-10-26
- Microsoft demos intent-classifying agents built with Copilot App and Foundry, deployable across Teams and Outlook — marlene_zw · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26