moe-runtime

Here is 1 public repository matching this topic...

SuperMarioYL / xpref

v0.2.0: live attach now measures real next-token recall (was always 0.000); every prediction is prefetched exactly once (was doubled); n-gram prior fixed from inert exact-set matching to per-expert transitions — sample recall@16 0.7405→0.8154; quickstart attach command fixed; version bump across all surfaces + CHANGELOG.

inference moe nvme llm deepseek ai-radar kimi-k3 expert-prefetch router-logit moe-runtime

Updated Sep 8, 2026
Python

Add this topic to your repo

To associate your repository with the moe-runtime topic, visit your repo's landing page and select "manage topics."

Learn more