1314521gjy/cyber-on-strata

在 8 GB 显存的显卡上跑 35B MoE（Cyber-Tiel-Coder-35B-A3B）：专家权重流式上卡、未命中专家由 CPU 池就地计算。Strata 引擎的 fork（MIT）。RTX 4080S 实测 8 GB 档 57.7–63.5 tok/s、全常驻 134–136 tok/s。Build from THIS repo — upstream does not support this model.

Language: C++

Observed star growth

Latest OpenPulse observation: 2026-10-19 02:03 UTC

GitHub topics

• cuda
• expert-offload
• gguf
• llm-inference
• low-vram
• moe
• qwen35moe
• speculative-decoding

View source repository on GitHub