wehub-resource-sync
26382a7ac6
CI / Clippy (push) Failing after 15m13s
CI / Test (ubuntu-latest) (push) Failing after 16m1s
CI / Test (macos-latest) (push) Has been cancelled
CI / Test (windows-latest) (push) Has been cancelled
CI / Build (no embeddings / no ORT) (push) Has been cancelled
CI / Format (push) Has been cancelled
CI / Cookbook (Node) (push) Has been cancelled
CI / Pi Extension (Node) (push) Has been cancelled
CI / Rust SDK (lean-ctx-client) (push) Has been cancelled
CI / Embed SDK (lean-ctx-sdk) (push) Has been cancelled
CI / Python SDK (leanctx) (push) Has been cancelled
CI / Hermes Plugin (Python) (push) Has been cancelled
CI / SDK Conformance Matrix (push) Has been cancelled
CI / Coverage (push) Has been cancelled
CI / cargo-deny (push) Has been cancelled
CI / Adversarial Safety (push) Has been cancelled
CI / Benchmarks (push) Has been cancelled
CI / Output-Quality Gate (eval A/B) (push) Has been cancelled
CI / Documentation (push) Has been cancelled
CI / CI Green (push) Has been cancelled
JetBrains Plugin / Actionlint (push) Has been cancelled
CodeQL / Analyze (actions) (push) Has been cancelled
CodeQL / Analyze (javascript-typescript) (push) Has been cancelled
CodeQL / Analyze (rust) (push) Has been cancelled
JetBrains Plugin / Validation (push) Has been cancelled
JetBrains Plugin / Build (push) Has been cancelled
JetBrains Plugin / Test (push) Has been cancelled
Security Check / Security Scan (push) Has been cancelled
1.8 KiB
1.8 KiB
Attention-aware Layout Driver v1 (AttentionLayoutDriverV1)
GitLab: #2311
LeanCTX kann Context nicht nur komprimieren, sondern auch re-layouten, damit relevante Teile an Positionen landen, die LLMs tatsächlich stärker beachten (“Lost in the Middle” → empirisch eher L‑Curve als U‑Curve).
Ziele
- Deterministisch: gleicher Input + gleiche Keywords + gleiche Policy ⇒ gleiche Reihenfolge.
- Semantic-first: bei größeren Inhalten zuerst Chunk‑Reordering (Imports/Types/Fns), danach line-level fallback.
- Policy-gated: reordering ist opt-in pro Profile.
- Verifier-safe: Reorder darf keinen Content “verlieren”, nur umsortieren.
- Bounded: kleine Inhalte werden nicht re-ordered (Edge Cases).
Aktivierung (Policy)
Per Profile:
profile.layout.enabled = true|falseprofile.layout.min_lines = <n>
Default: enabled=false.
Semantik (v1)
- Small files: wenn
lines <= 5(oder< min_lines) ⇒ keine Änderung. - Large content: ab
lines >= 15wird Chunking versucht:detect_chunks(content)order_for_attention(chunks, task_keywords)render_with_bridges(ordered)
- Fallback: sonst line-level scoring + stable tie-break (original index).
Keywords
Keywords stammen aus dem Task/Intent Kontext (z.B. task Argument), extrahiert via task_relevance::parse_task_hints.
Determinism Guarantees
- Sorts haben stabile Tie-breaks (z.B.
start_line,original_index), damit gleiche Scores nicht zu nondeterministic reorder führen.
Relevanter Code
- Driver:
rust/src/core/attention_layout_driver.rs - Chunk reorder:
rust/src/core/semantic_chunks.rs - Line-level reorder:
rust/src/core/neural/context_reorder.rs - Learned attention curve:
rust/src/core/neural/attention_learned.rs