Technical deep-dives on the Mostlysane AI research stack — coming soon.
Coming Soon
This page will document:
- Ternary (1.58-bit) quantization port and MoE integration
- TurboQuant KV cache — turbo3, turbo4, turbo6 type benchmarks
- Entropy-adaptive KV cache — per-layer precision profiling (Path B)
- CodeNeedle evaluation results across model/config combinations
- Sparse V dequant research (GPU accelerated)
In the meantime, see our visual model comparisons or get started.