Tuning Qwen3 27B as a coding agent on 2x3090s cuts turn latency from 28s to 7s
bolts98 · reddit · 2026-09-19
A Reddit user shares how they tuned Qwen3 27B with vLLM + Oh My Pi (OMP) as a coding agent on 2×RTX 3090s, cutting average wait per turn from 28s to 7s. Key changes:
- Explicit effort level on every role (unset ones defaulted to xhigh)
- thinkingtokenbudget of 7500
- maxTokens 8k → 32k (file writes were getting cut off)
- Tool output over 10KB redirected to a file
- Max 4 subagents, appendOnlyContext enabled
More from coding & agent
- Hindsight ships 0.8B model for agent memory decisions: 2.5x cheaper, 1.9x faster — victor_explore · 2026-11-04
- Human Oversight Is 70-75% of AI Agent Costs in Banking, McKinsey Finds — mikeflache · 2026-11-04
- Over 80% of Major Financial Institutions Now Run Agentic AI in Production — mikeflache · 2026-11-04
- Mnemos MCP nears launch after shared memory layer fix — RileyRalmuto · 2026-11-04
- Moving mobile dev to remote machines makes it multiplayer across product, design, engineering — rudrank · 2026-11-04
- Kuaishou's EvoRec: self-evolving agentic recommender lifts ad revenue 4.6% — _reachsumit · 2026-11-04