ByteShape ships full Qwen 3.8 27B quants: 3.84bpw hits 99.63% of BF16, KLD debunked as quality metric
enrique-byteshape · reddit · 2026-09-15
ByteShape released full ShapeLearn GGUF quants of Qwen 3.8 27B with benchmarks across six GPUs (RTX 6000 Pro Blackwell down to 5060 Ti):
- Accuracy: 3.84bpw (GPU-5) reaches 99.63% of BF16's aggregate score over 8 benchmarks; 3.23bpw still hits 98.72%. All five quants sit on the quality/speed-bpw frontier vs Unsloth v3, ISTA-DASLab, AtomicChat and Bartowski.
- KLD isn't a leaderboard: Unsloth's UD-IQ3S has 20% lower KLD than a similar-size Lite model yet scores worse (95.55% vs 97.33%). KLD only catches broken quants — the thesis of their EMNLP 2026 Industry Track paper.
- Speed: DFlash2 gives 1.34–2.10× throughput; MTP 1.28–1.66× with temperature sampling.
- Benchmarks cover GSM8K, MMLU, IFEval, LiveCodeBench V6, Multi-IF, ACEBench, Multiple HumanEval and BFCL V4 across instruct and thinking modes.
More from Infra
- To dethrone Postgres, new databases must prioritize ecosystem integrations from day one — mgill25 · 2026-11-02
- ESP32-powered 4-inch touchscreen builds a TUI dashboard for monitoring a DGX inference cluster — CATLLM · 2026-11-02
- OpenRouter crosses 10 quadrillion annualized token run rate, doubling every 10.5 weeks — deedydas · 2026-11-02
- Palmer Luckey asks: can you use datacenter RAM for things prohibited by the operator's terms of service? — avaitopiper · 2026-11-02
- Disable MTP for 20-30% faster local LLM inference on limited RAM — InternalMode8159 · 2026-11-02
- Kioxia up 216% in a year but down 68% from its June 2026 peak — zephyr_z9 · 2026-11-02