eviction

REAL: REtrieval-reAsoning and Logic-constructed Attention Behaviors for Long-Context KV Cache Compression

Custom CUDA kernels for KV-cache eviction + INT8 quantized paged attention (vLLM-oriented), PyTorch C++ extensions with Python API: eviction @16k blocks 1707.75us host -> 562us fused (~3x); INT8 attention p50 1.76-14.18ms; ~49.9% KV memory vs FP16; 26/26 pytest passing on T1000.

Which Redis eviction policy keeps the most useful data, at what memory budget, for what shape of traffic - measured against Belady's optimal, the best any policy could have done

Analytical benchmark for sliding window attention KV cache management: quality vs window size tradeoffs, SWA vs eviction comparison, prefix sharing interaction, and operational window recommendations across four attention distributions

Calibrated simulation benchmark for KV-cache compression pipeline ordering in LLM serving, comparing quantization, eviction, tiering, prefetch, and reload trade-offs.

Agent memory with eviction that is not 'keep everything'. Episodic vs semantic, contradictions resolved on write, and a report showing recency loses 71% of the lookups your agent actually makes. Zero deps.

Vorn: Residual Direction, Familial Eviction, and the Granularity Rescue Spectrum — paper + reproducible code, data, and figures (Zenodo DOI: 10.5281/zenodo.20519215)

Predictive model (R, Negative Binomial regression) forecasting monthly eviction filings at census tract level for Philadelphia. Python pipeline handles fairness auditing, PDF reporting, dashboard, and resident notification. Built for resource allocation, not automated enforcement.