Inference

Writing

• How to Optimize a CUDA Matmul Kernel for cuBLAS-like Performance: a Worklog
• Taking vLLM Apart: A Practical Guide to Disaggregated Serving

Servers

• ninfer: RTX 5090-optimised inference
• Strata: Qwen3.8-Flash-Next through intelligent offloading
  • PolyStrata: Strata fork supporting more models
  • Project Maya: Strata-based runner for GLM-5.3-Flash / DeepSeek-V4.1-Flash-1017