← sources

I need context window

source: reddit (reddit:LocalLLaMA)published 2026-11-13 14:33ingested 2026-11-14 02:42

retained text

I started developing a full fledged application using Chatgpt plus and VScode+GitHub copilot (local llm)

I used Astra max chat for overall planning, and asked it to put a starter kit on connect gdrive and create prompts such that local LLM can work on small tasks using an agent orchestrator and sub agents with entry and exit criteria, tests for every small task.

I have a 5090 (+ 128gb ddr5) rig running Qwen 3.8 27b NVFP4 at default settings with 104k context window. 

Problem:

The local rig is able to start on tasks but taking a huge amount of time and is running only one subagent at a time. Been almost 24 hrs.

I have 2x 3090 +128gn ddr4 as well, currently not being used. 3.8 FN runs at about 90tps

Anything better I can do here?

document fetched 2026-11-14 00:46 · canonical https://reddit.com/r/LocalLLaMA/comments/1xv9ga2/i_need_context_window