98c96cda6f
All checks were successful
Smoke Test / smoke (pull_request) Successful in 9s
Adds Emscripten build support and a browser-based roundtrip demo for TurboQuant's polar_quant_encode_turbo4 / polar_quant_decode_turbo4 functions. Changes: - CMakeLists.txt: detect Emscripten toolchain; build shared WASM module with exported C functions; set ALLOW_MEMORY_GROWTH and MODULARIZE - Added build-wasm.sh: convenience script using emcmake to produce libturboquant-wasm.js + libturboquant-wasm.wasm - Added wasm-demo/: self-contained HTML/JS demo that loads the WASM module and runs an encode/decode roundtrip displaying latency This establishes the build pipeline and client-side harness needed to run TurboQuant in the browser. A working demo is now one `./build-wasm.sh` plus `python3 -m http.server` away. Closes #104
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
- Project Status — Full project status and build specification