Tencent Open-Sources STEPQuant, Cutting Inference Memory by 68.7%

Tencent's research team open-sourced STEPQuant, a spatial-temporal post-training quantization framework that quantizes Delta-rule recurrent states to 6-bit, cutting serving memory by up to 68.7% without accuracy loss.

2026-10-08 ~ 2026-10-09 · 2 related posts

• STEPQuant: 6-bit quantization of Delta-rule recurrent states cuts serving memory by up to 68.7% — zju-community · 2026-10-08
• Tencent's STEPQuant: 6-bit recurrent states match FP32 with 68.7% less memory — _akhaliq · 2026-10-09