Skip to content

Latest commit

 

History

History
20 lines (17 loc) · 1.53 KB

File metadata and controls

20 lines (17 loc) · 1.53 KB

Changelog / Milestones

Major milestones from the v1 development cycle. The optimization history tables are in benchmarks.md and speed-experiments.md.

v1 Complete, Speed Optimized, Community-Tested

  • 511+ Python tests, 100% code coverage on diagnostics
  • C port integrated into llama.cpp with Metal GPU kernels
  • --cache-type-k turbo3 --cache-type-v turbo3 works on Apple Silicon (turbo2/turbo3/turbo4 all supported)
  • turbo2 Metal support: 2-bit, 6.4x compression, +6.48% PPL — for extreme memory pressure or asymmetric K/V
  • q8_0 prefill speed parity achieved (2747 vs 2694 tok/s)
  • Norm correction: PPL beats q8_0 on CUDA (-1.17%), +1.1% on Metal (ported from @spiritbuun)
  • 4-mag LUT: auto-detected on M1/M2/M3/M4, +38-45% decode at long context
  • Layer-adaptive mode 2: q8_0 quality at 3.5x compression (last 8 layers at q8_0)
  • Temporal decay: 30-34% memory savings at long context (experiment branch)
  • NIAH retrieval: 9/9 single needle with sparse V (vs 7/9 baseline), 100% multi-key through 32K. 30/30 on Llama-70B, 10/10 on Command-R+ 104B
  • 14 decode approaches tested on M2 Pro — comprehensive hardware analysis
  • Stress tested up to 104B: Command-R+ 104B Q4_K_M at 128K context (PPL 4.024). Llama-70B Q4_K_M at 48K (PPL 4.019). turbo3 prefill faster than q8_0 at 32K on both models
  • Community: 30+ testers across M1/M2/M3/M5 Mac, RTX 3080 Ti/3090/4090/5090, AMD 6800 XT/9070 XT
  • Rotation Gaussianization validated on real Qwen3 KV tensors (kurtosis 900 → 2.9)