+
+
A local, MoE-offload inference runtime with an OpenAI- and Anthropic-compatible
HTTP API — Run DeepSeek-V4-Flash on your 5090 with 20+ TPS.
-[](https://join.slack.com/t/flashml/shared_invite/zt-3zpdh5j10-9dwTXrgLiqpVxizhA9KVbA)
-[](https://discord.gg/xzwSnMdsX)
## Quick start