410a0a56c0
All checks were successful
Smoke Test / smoke (pull_request) Successful in 28s
Security improvements: - Input validation (dimension must be power of 2, <= 4096) - Null pointer checks for all parameters - Constant-time quantization (no data-dependent branches) - Bounds checking in bit packing/unpacking - Safe wrapper functions (safe_polar_quant_encode/decode_turbo4) - RAII SafeBuffer for memory safety Added turbo-safety.h with: - is_power_of_2() validation - validate_dimension() with clear error messages - validate_pointers() for null checks - ct_abs(), ct_min_index(), ct_abs_diff() for constant-time ops - SafeBuffer<T> RAII wrapper Updated llama-turbo.cpp to use validation and constant-time ops. Updated llama-turbo.h with safety documentation. 13 tests pass. Fixes #55
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
- Strago: Build spec author
- Cid: Implementation, benchmarks, deployment
- Locke: Research support, upstream watch
- John: Quality review
- Frankie: Coordination
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
- BUILD-SPEC.md — Full build specification (Strago, v2.2)