Yamz Labs

Kyojin, our open-source engine, runs the largest open models on one mini PC or one graphics card, about 2× faster than llama.cpp (up to 3× on some models). AMD (ROCm) and NVIDIA (CUDA), built on ExLlamaV3.

210 GitHub stars in 3 weeks · ~1,000 developers installed it · 22 releases

Here we publish the compressed models it runs: quality checked against the official weights, ready to download and serve.

• GLM-5.3-Flash
• Qwen3.8-Flash-Next
• MiMo-V2.6-Flash
• DeepSeek-V4.1-Flash (experimental)
• Nanbeige4.3

Engine: Kyojin on GitHub · Website: yamzlabs.com · News: @YamzLabs on X

Not affiliated with the original model authors.