Zhejiang University

Trends

• Efficiency methods cut the compute and memory models need
• Agents improve themselves through skills, harnesses, and memory

Items

1. Aug 28

   Zhejiang University's label-free TTPO

   A test-time training method that distills rollouts agreeing with a majority-vote pseudo-label while penalizing confident disagreements via grouped RL, matching label-supervised self-distillation and lifting Qwen3-1.7B from 38.0% to 45.2% on competition math.

2. Jun 15

   Zhejiang University's TokenPilot

   A dual-granularity context manager that stabilizes prompt prefixes and evicts stale context only when its task relevance expires, cutting long-horizon agent inference costs by 56-87% without breaking the prompt cache.

3. Apr 6

   TriAttention compresses the KV cache 10.7x

   A training-free method from MIT, NVIDIA, and ZJU scores KV importance from the stable trigonometric structure of pre-RoPE queries and keys, matching full attention on AIME25 at 2.5x throughput or 10.7x less KV memory and fitting a 32B reasoning model on a 24GB RTX 4090.