Hy4-preview quantized to 200GiB with minimal accuracy loss
TencentHunyuan · x · 2026-08-30
Tencent's Hunyuan Hy4-preview (770B MoE) has been compressed from 1.5TB to 200GiB GGUF using the MIX-STQ10 method, achieving a 7x reduction. Instead of uniform low-bit quantization, the technique treats bit-width as an allocation problem, using calibration data to assign specific precision (1.31-bit to 2.06-bit) per layer. Benchmarks like MCP Atlas and SWE-Bench show minimal performance collapse compared to the BF16 baseline.
More from Infra
- engy.ai processes 50B+ tokens in a day as open-model inference demand surges — const_reborn · 2026-10-16
- New inference engine 'e' hits 10,500 tok/s on a single DGX Station — brandon_galang · 2026-10-16
- SGLang ships Day-0 support for FLUX 3 Image Klein: 10 reference images, BF16/FP8/NVFP4 variants — ying11231 · 2026-10-16
- NVIDIA's Nemotron 3.5 Super VL hits CoreWeave: 12B-active multimodal MoE at $0.25/M input tokens — wandb · 2026-10-16
- Chip equipment beats bull case: 2026 WFE seen at $153B, $278B by 2030 — BenBajarin · 2026-10-16
- Steam now launches in 5.23s, beating Epic's 5.47s — and vLLM gets roasted for slow startup — AlpinDale · 2026-10-16