Tencent Hunyuan releases lightweight model with 85% weight compression
智东西 · wechat · 2026-09-01
Tencent released a lightweight version of its flagship Hunyuan Hy4preview model, compressing weights from 1.5TB to 214GB using proprietary Sherry sparse ternary quantization and MIX-STQ1.0 mixed precision strategies. Tests show the compressed model performs on par with the original in long-context understanding and retrieval, with a slight drop in math. Using the distributed inference framework prima.cpp, the team demonstrated running the model across heterogeneous devices (e.g., a 4090 laptop and A4000 server), achieving 6x faster inference than single-machine offload.
Related event: Tencent Hunyuan's AngelSlim Shrinks Hy4-preview from 1.5TB to 214GB(5 posts)→
More from Infra
- Auction-based LLM routing paper finds fixed token pricing wastes 10%-71% in costs — _onionesque · 2026-10-18
- Running Ling 3.1 Flash (560B MoE, 25B Active) Across 4 DGX Sparks via llama.cpp RPC — BerryBlush_Rogers691 · 2026-10-18
- SGLang Summit sponsors jump from 11 to 36 in one week — ying11231 · 2026-10-18
- Oracle CDS hits record 300bps, called first AI trade domino after Barclays sell downgrade — arthurcolle · 2026-10-18
- Agent Coding Workflow Would Cost ~$40k/Month on API Pricing, Dev Reveals — ChrisRackauckas · 2026-10-18
- PS5LM: llama.cpp port turns a jailbroken PS5 into a home LLM server (500+ GitHub stars) — wateriscoding · 2026-10-18