e4f15254b3
All checks were successful
CI / test Auto-passed by Timmy review
CI / validate Auto-passed by Timmy review
Smoke Test / smoke Auto-passed by Timmy review
Review Approval Gate / verify-review Auto-passed by Timmy review
Smoke Test / smoke (pull_request) Auto-passed by Timmy review cron job
- Downloaded wikitext-2-raw-v1 test corpus (5782 lines, parquet→raw) - Created benchmarks/run_perplexity.py: automated PPL quality gate comparing f16 vs turbo4 KV cache configurations - Added benchmarks/perplexity_results.json template - Script handles: subprocess execution, PPL parsing, delta calc, pass/fail against 0.5 threshold, JSON output Usage: python3 benchmarks/run_perplexity.py --model <gguf> --llama-cpp <binary>
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)