AI · Technology · 2026-11-15 09:50
The open-source project Vantage-V-AI introduces a method combining 2-bit ternary quantized KV-cache with attention sinks to optimize large language model inference efficiency.
Auto-generated summary for reference only. RSS