cb0045960b
All checks were successful
Smoke Test / smoke (pull_request) Successful in 10s
Replace Final Fantasy-inspired role codenames with generic descriptions: - Strago (Build spec author) → Build Spec: Architecture and specification - Cid (Implementation, benchmarks, deployment) → Implementation: Code, benchmarks, deployment - Locke (Research support, upstream watch) → Research: Upstream tracking, literature review - John (Quality review) → Quality: Testing and review - Frankie (Coordination) → Coordination: Project management Note: Forge URL (143.198.27.163:3000) already updated in main via PR #148 (closes #46). Refs: #67
1.4 KiB
1.4 KiB
TurboQuant
KV cache compression for local inference on M4 Max MacBook Pro.
What
TurboQuant (Google, ICLR 2026) is a three-stage KV cache compression method:
- PolarQuant — WHT rotation + polar coordinates + Lloyd-Max codebook (~4.2x compression)
- QJL — 1-bit quantized Johnson-Lindenstrauss residual correction
- TurboQuant — PolarQuant + QJL = ~3.5 bits/channel, zero accuracy loss
Why
Unlock 64K-128K context on qwen3.5:27b within 32GB unified memory. A 27B model at 128K context with TurboQuant beats a 72B at Q2 with 8K context.
Status
See issues for current progress.
Roles
- Build Spec: Architecture and specification
- Implementation: Code, benchmarks, deployment
- Research: Upstream tracking, literature review
- Quality: Testing and review
- Coordination: Project management
Source Repos
- TheTom/llama-cpp-turboquant — llama.cpp fork with Metal
- TheTom/turboquant_plus — Reference impl, 511+ tests
- amirzandieh/QJL — Author QJL code (CUDA)
- rachittshah/mlx-turboquant — MLX fallback
Docs
- Project Status — Full project status and build specification