Developer builds Shura: a local coding assistant on one RX 7800 XT hitting 78 tok/s that should improve with use
HIKKIAN · reddit · 2026-11-09
A Reddit developer detailed Shura, a local coding assistant on a single RX 7800 XT (16GB) running a Qwen3.6-35B-A3B MoE finetune: 78 tok/s decode with MTP, 200k context, 1,000 tok/s prefill on 100k input, graceful VRAM handback to the desktop in 0.25s. The roadmap: sourced memory, test-gated rules, test-driven multi-sample compute, and — the unproven part — LoRA adapters trained on confirmed examples atop the same quantized base, with all results to be published either way.
More from coding & agent
- Personal agents can't handle humans' comfortable 'logistical chaos' — builderjaydub · 2026-11-09
- kitlangton proposes content-addressed compilation units so coding agents can share work across machines — kitlangton · 2026-11-09
- Fixed memory rules hurt strong agents: "record only failures" costs Opus 5 gains — curious_vii · 2026-11-09
- Building a language for agents: Blood makes regex and SQL typos compile-time errors — kitlangton · 2026-11-09
- macOS 27.3 lets Mac VMs use host GPU and Neural Engine, up to 7× faster Core AI — SumitGup · 2026-11-09
- McKinsey: Top 20-25% of Companies See 2x AI Coding Gains — Here's What They Do Differently — AI Engineer · 2026-11-09