Local Inference 1 Self-Hosted AI Coding Agent on a Tesla V100 Server GPU — VRAM Optimization & Speculative Decoding Jun 18, 2026