27b

vLLM Qwen 3.6-27B (AWQ-INT4) + DFlash speculative decoding on AMD Strix Halo (gfx1151 iGPU, 128 GB UMA, ROCm 7.13). 24.8 t/s single-stream, vision, tool calling, 256K context, OpenAI-compatible, Docker. Matches DGX Spark FP8+DFlash+MTP at a third of the cost. No CUDA.

Run Qwen 3.6-27B AWQ-INT4 models with DFlash speculative decoding on AMD Strix Halo hardware using vLLM for high-throughput inference.

local-c1: a DSH helper that registers a local Qwen (qwen3.8-27b) model on a 3090 node via one idempotent Node script—patches settings.yaml and the dual preset's compaction policy. Inference served by Don-Chad/ninfer-3090. Supports dry-run and auto-backup; restart DSH to apply.

local-c8-worker: a DSH dual-model toolkit pairing a DeepSeek brain with a local Qwen3.8-27B worker (8192-token context, served by Don-Chad/ninfer-3090). Ships a deployer, presets, skills, and a fallback client; splits tasks into chunks, fans out, then aggregates and compacts.