GitHub

perkel666/MegaCapybara

The fastest inference engine for Qwen3.8-27B on the NVIDIA RTX 5090: up to 500 tokens/s for one agent and up to 2,000 tokens/s for many, contexts up to 1M tokens, and a launcher that shows what every setting costs. Windows and Linux.

Stars
52
Forks
1
Stars gained in 24 hours
0
Forks gained in 24 hours
0

Topics: anthropic-api, claude-code, cuda, inference-engine, linux, llm, local-llm, openai-api, qwen, rtx-5090

Latest data point: Nov 21, 2026 · 52 stars · 1 forks

Open on GitHub