AI Trends

Trending story

RecSA Achieves Extrapolation to 2M Tokens with Logarithmic Query Cost

1 reports1 story updatesUpdated Nov 19 03:20

Understand this story

Story summary

The English summary is being updated. See the reporting timeline below.

Latest development · Report

RecSA Extrapolates 32K Training to 2M Tokens at Logarithmic Cost

Developments

1 confirmed developments

  1. English translation is being prepared.

    Source:UCSD Hao AI Lab

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 19

  1. UCSD Hao AI LabFirst-party

    RecSA Extrapolates 32K Training to 2M Tokens at Logarithmic Cost

    UCSD Hao AI Lab introduced RecSA, which builds a tree over the KV cache and searches it level by level, making the cost of each query scale as O(log L) and enabling a model trained only on 32K sequences to extrapolate to 2M tokens. After training Qwen3-4B on 0.62B tokens, RecSA outperformed all baselines on RULER at lengths from 131K to 2M, scoring 60.5 at 1M versus 40.0 for the strongest baseline, Dense + DCA. On a single H200, RecSA delivered a 3.2× prefill speedup and a 3.8× decoding speedup at 524K. The code includes Triton and CUDA kernels, Qwen3 integration, and evaluation tools.

Related stories

  1. Related storiesRecSA Outperforms All Baselines in the RULER Long-Context Benchmark
  2. Related storiesRecSA Extrapolates Context to 2 Million Tokens with Logarithmic Query Cost
  3. Related storiesRecSA Enables Models Trained on 32K Tokens to Extrapolate to 2M Tokens
  4. Related storiesRecSA Achieves Logarithmic Query Costs with a KV Cache Tree