Apple-Silicon
- Sub-Bit Ternary Quantization on Metal, Speculative Tool Execution for MCP, and Cloudflare's Post-Quantum TLS 1.3 Rollout
1.58-bit ternary weight packing unlocks 70B model inference at 65 tokens/sec on Apple Silicon Metal shaders, while speculative MCP tool execution cuts agent round-trip latency in half. Issue #115.