03cf42e29d
All checks were successful
Smoke Test / smoke (pull_request) Successful in 11s
Add compare_configs.py — orchestrates running 4 Gemma4 configs in parallel and produces side-by-side comparison table with aggregated metrics (TTFT, tokens/sec, latency, peak memory). Picks winner by highest tokens/sec. Configurations: 1. Ollama gemma4 (baseline) 2. llama-server gemma4 f16 KV 3. llama-server gemma4 turbo4 KV 4. llama-server gemma4 turbo4 + layer-adaptive Also adds comprehensive test suite (13 tests) covering ConfigEntry, aggregation, table building, demo mode, and prompt loading. Closes #29 Co-authored-by: step35-cli <step35-cli@timmy.foundation>
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