Language models learn to evict KV cache entries end-to-end via reinforcement learning from outcome reward alone, achieving 2-3x cache compression while maintaining accuracy on Countdown, AMC, and AIME tasks.
The subtlety is that g is emitted deterministically at inference, so its probability is degenerate and admits no sampled log-probability
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Neural Garbage Collection: Learning to Forget while Learning to Reason
Language models learn to evict KV cache entries end-to-end via reinforcement learning from outcome reward alone, achieving 2-3x cache compression while maintaining accuracy on Countdown, AMC, and AIME tasks.