• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
  • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI
    AI

    alphaXiv Shares Paper on Probabilistic KV Cache Eviction

    Tweet from alphaXiv account describes a paper reframing KV eviction with sampling and weighting.

    AL
    1 Source, 30d ago, first seen 30d ago

    TLDR

    The official alphaXiv account posted about the paper “A Probabilistic Interpretation of KV Cache Eviction.” The post states that most KV eviction methods rely on top-k heuristics, which can distort attention after tokens are removed. It says the paper instead treats eviction as expectation estimation and applies probabilistic sampling plus importance weighting to correct the distortion. The tweet includes a static screenshot of the paper’s first page. No other details or confirmations appear in the packet.

    Combined views

    11.5K

    1 Source, first seen 30d ago

    Combined views

    11.5K

    1 Source, first seen 30d ago

    179 likes
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet

    179 likes
    2 comments
    128 saves
    27 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    2 comments
    128 saves
    27 reposts

    Sentiment

    Positive——Negative

    Summary

    Not enough discussion yet.

    No sentiment analysis available yet.

    1 Source

    @askalphaxiv“A Probabilistic Interpretation of KV Cache Eviction” Most KV eviction methods use top-k heuristics to decide what to keep, but this can heavily distort attention after tokens are removed. This paper reframes KV eviction as expectation estimation, then uses probabilistic sampling and importance weighting to correct for missing KV entries during decoding. This gives a principled bias-variance tradeoff that stays competitive with existing methods while being more robust. https://www.alphaxiv.org/abs/2608.28293

    1 Source

    @askalphaxiv“A Probabilistic Interpretation of KV Cache Eviction” Most KV eviction methods use top-k heuristics to decide what to keep, but this can heavily distort attention after tokens are removed. This paper reframes KV eviction as expectation estimation, then uses probabilistic sampling and importance weighting to correct for missing KV entries during decoding. This gives a principled bias-variance tradeoff that stays competitive with existing methods while being more robust. https://www.alphaxiv.org/abs/2608.28293