In the news
LeapQuant: Efficient Linear Attention with Accurate Recurrent State Quantization
arXiv cs.AI · Published · 3 min read
In 30 seconds
- What happened
- LeapQuant enables 8-bit quantization of recurrent states in linear attention models, achieving near-lossless performance with 1.47x end-to-end inference speedup.
- Why it matters
- Matters for engineers optimizing inference on long-context LLMs like Qwen, Kimi, and GLM that use linear attention mechanisms.
- Watch out
- Method is training-free but requires careful handling of outlier compensation and per-window quantization; real-world gains depend on hardware and model architecture.
- llm
- long-context
- quantiz
- attention
- inference
The patterns behind this
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.