[GitHub Trending] JustVugg/colibri

8.3 relevance

Run large MoE models on local hardware with zero deps, innovative and relevant.

AI/ML github.com

Answer the board on AI readiness

Arrive with evidence, not opinion. See how your foundation benchmarks against your peers, get a clear read, and leave with a ranked place to start.

Why It Matters

For a solutions architect focused on AI/ML and infrastructure, Colibrì demonstrates how aggressive systems research at the software/hardware boundary can reduce hardware dependency and inference cost, enabling large models on existing hardware.

Key Takeaways

Evaluate Colibrì's multitiered inference approach to run large open-source MoE models on your current infrastructure and reduce cloud spend.

Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦 - JustVugg/colibri

Summary

Colibrì is a pure-C open-source inference engine that runs 744B–2.8T MoE models on consumer hardware via a unified VRAM/RAM/disk hierarchy, supporting nine model families including GLM-5.2, Kimi K3, and DeepSeek V4. It provides a live web dashboard and expert cortex visualization, and guarantees model semantics are preserved even with limited fast memory, making frontier models accessible without proprietary infrastructure.

Author

JustVugg