core-deepinfer , A readable NumPy inference loop: prefill, prefix recovery...

core-deepinfer is a NumPy‑based inference loop that supports prefill, prefix recovery, and speculative KV rollback. The lightweight Python code offers a readable alternative to GPU‑heavy frameworks, enabling quick experimentation with KV‑cache and speculative decoding techniques. It has attracted 10 GitHub stars since launch, signaling growing interest in CPU‑friendly LLM inference.

Opening Kapyn…