github-featured / yvgude--lean-ctx

发布

0 发布 257 Git 标签

RSS 订阅源
  • v2.3.0 8d225aacd1
    比较

    release: v2.3.0 — Scientific Compression Engine
    Release / Build aarch64-apple-darwin (push) Has been cancelled
    详情
    Release / Build x86_64-apple-darwin (push) Has been cancelled
    详情
    Release / Build aarch64-unknown-linux-gnu (push) Has been cancelled
    详情
    Release / Build x86_64-unknown-linux-gnu (push) Has been cancelled
    详情
    Release / Build x86_64-pc-windows-msvc (push) Has been cancelled
    详情
    Release / Create Release (push) Has been cancelled
    详情

    frostbyte_neo 发布于 2026-03-26 16:36:21 +00:00

    10 information-theoretic optimizations derived from Shannon entropy,
    Kolmogorov complexity, Bayesian inference, and rate-distortion theory:

    • BPE token-aware entropy filtering (I1)
    • N-gram Jaccard + Minhash deduplication (I2)
    • Cross-file dedup with block references (I3)
    • Bayesian mode predictor with persistence (I4)
    • Adaptive LITM profiles per LLM model (I5)
    • Boltzmann cache eviction with token budget (I6)
    • Information density metric in quality scoring (I7)
    • Auto-delta encoding for changed files (I8)
    • Huffman instruction templates (I9)
    • Kolmogorov complexity proxy for mode guidance (I10)

    Session benchmark: 69% total token savings (149K → 46K tokens).
    Cache re-reads: 99%, Map mode: 97.6%, Auto-delta: 98.9%.

    Also: lib.rs crate restructure, Default impls, clippy clean.
    Made-with: Cursor

    下载附件