Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
C
docs: acknowledge inference research foundations
ZacharyZcR99cd64007ec8
diff --git a/README.md b/README.mdindex c450ebac..4936f103 100644--- a/README.md+++ b/README.md@@ -520,6 +520,30 @@ releasing frontier-class weights in the open — **Z.ai** (GLM), **Moonshot AI**contributor who benchmarked, bisected, replicated an atlas run, or sent a patch.This project is proof of what open weights make possible.+The project's expert placement, compression, and routing experiments also build+on ideas and evidence from the following open research and systems work:++- [REAP](https://github.com/CerebrasResearch/reap) and+ [EASY-EP](https://github.com/RUCAIBox/EASYEP) for output-aware and+ domain-specific expert importance.+- [SERE](https://github.com/JL-Cheng/SERE) for similarity-based expert+ re-routing, and [ReMoE](https://github.com/BUAA-OSCAR/ReMoE) for+ cache-locality-aware router fine-tuning.+- [MC-SMoE](https://github.com/UNITES-Lab/MC-SMoE) for routing-guided expert+ merging and compression.+- [MoBE](https://github.com/inclusionAI/MoBE) and+ [D²-MoE](https://github.com/lliai/D2MoE) for shared expert bases and+ low-rank expert deltas.+- [HybriMoE](https://github.com/PKU-SEC-Lab/HybriMoE) for hybrid CPU/GPU expert+ scheduling, [ScMoE](https://arxiv.org/abs/2404.05019) for overlapping expert+ communication with computation, and+ [OD-MoE](https://arxiv.org/abs/2512.03927) for distributed on-demand expert+ loading.+- [vLLM](https://github.com/vllm-project/vllm),+ [llama.cpp](https://github.com/ggml-org/llama.cpp), and+ [kTransformers](https://github.com/kvcache-ai/ktransformers) for the open+ inference systems and expert-offload work that make comparisons reproducible.+## LicenseApache 2.0. GLM-5.2 weights are released by Z.ai under MIT.