NVIDIA Integrates TriAttention: Trigonometric KV Compression for Long Context
青稞AI · wechat · 2026-08-20
To address the VRAM bottleneck of KV Cache in long reasoning, NVIDIA researchers proposed TriAttention. The method leverages the observation that pre-RoPE Q/K vectors cluster tightly around a fixed center. By approximating attention logits with trigonometric series based on this centrality, TriAttention effectively scores key importance for cache pruning. It has been merged into NVIDIA TensorRT-LLM and adopted by the LongLive video generation framework, reducing local-attention KV memory by 50% without quality loss.
More from Infra
- Haight Street Compute Market: a one-day trading floor for GPU deals in SF — ctjlewis · 2026-10-16
- Temporal Serverless Workers on Cloud Run: queue-driven autoscaling, scale to zero — rseroter · 2026-10-16
- LMA brings day-0 model support to TPU for the first time, spanning NVIDIA, AMD, Intel via SGLang — ying11231 · 2026-10-16
- Semtech's analyst-day slide cleanly explains the four signal-integrity impairments — BenBajarin · 2026-10-16
- Fortytwo Engine boosts Qwen3.8-27B accuracy 7% and cuts tokens 25% via Swarm Inference — rohanpaul_ai · 2026-10-16
- Berkeley scholars: concentrated chip supply chain makes a hardwired frontier AI pause feasible — wfithian · 2026-10-16