If you want to reproduce my current local Hermes Agent + Qwen3.6-27B setup, this is the shape I would start from.
Target
One local coding agent.
One 24GB GPU.
Long context.
Qwen3.6-27B + vLLM + Hermes on 24GB VRAM: May 2026 Recipe
If you want to reproduce my current local Hermes Agent + Qwen3.6-27B setup, this is the shape I would...
TL;DRAI
Qwen3.6-27B on 24GB VRAM: vLLM configured with 131k context, prefix caching, and max_num_seqs=1; Hermes disables child agents and speculative decoding for stable multi-turn inference. The setup targets multi-hour agent sessions over throughput peaks—a practical trade-off for local coding agents avoiding KV-cache thrashing and OOM on consumer GPUs.
927 words~4 min read
