perkel666 / MegaCapybara
git clone https://github.com/perkel666/MegaCapybara.git
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.
+0
stars, 24h · #34,308
+~0
stars, 7 days · #34,783
+37
stars, 30 days · #13,248
41
★ total
1
forks · 3 open issues
Star history (as recorded here)
LicenseMIT
Created2026-10-03 (19d ago)
Last push1d ago
OwnerUser