5bfb9eb000
All checks were successful
Smoke Test / smoke (pull_request) Successful in 14s
benchmarks/compare_configs.py: - Runs 4 configs (ollama, llama-f16, llama-turbo4, llama-turbo4-adaptive) - Aggregates TTFT, tok/s, latency, peak memory - Picks winner by highest tok/s - Outputs JSON report + human-readable table - --demo mode for testing without live servers tests/test_compare_configs.py (13 tests): - ConfigEntry, ConfigResult, default configs - Aggregation logic, winner selection, table format - Demo mode with and without output file - Prompt loading from test_prompts.json Closes #29.
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)