Repository navigation
Releases
Releases · iscvlab/xlaya
v2.3.0 — Regime-Invariant 8D Code-Graph Sieve, Conformal ToolGate & Gemma 4 ADK Bundles
What's New in XLaya v2.3.0
Released alongside our Google Gemma 4 Developer Agent Competition Paper Track Writeup (Submitted!).
Core Architecture Upgrades (--profile nano • Zero PyTorch / Zero NumPy)
- 2-Tier Source-First Lexical Pre-Ranker & Compact AST Skeleton Headers: Partitions
.py/.pyirepositories intoTier-1core package files ahead ofTier-2docs_src/,tests/,examples/, andscripts/(0/129core gold truncation onKaggle 129), attaching compact enclosing class/import headers (100%split-method coverage). - 4-PC
.npzWhitening & Isolated Call-Graph Channels (F-01..F-08Regime Shield): Eliminates narrow-cone embedding anisotropy via All-but-the-Mean + Top-4 PC whitening (raw cosine median 0.8518 → -0.0158across127/127Kaggle archives) and isolates237,386matched NetworkXnode_link_dataedges without inflating AST degree (0degree cliff). - Smooth Channel Availability (
M_avail,γ_max = 1.45): Preserves calibrated reliability acrossRegime R1(with.json/.npzsidecars) and pure-Python stdlibRegime R2(|ΔS| = 0.0041,|ΔTop-1 ECE| = 0.0073,"torch" not in sys.modules,10.414 MBΔRSS,3.99 mswarm query /255.33 ms/repouncontended cold scan). - 2-Stage Causal Monotonic Tool-Schema Gate (
triage_tool_call): Split-conformal abstention & tool routing inpolicies/sieve_genomes.json(5,679 BLF /6,020 BCRLF), evolved strictly on125repo-disjoint@base_committasks (21unseen repos,1.28MCUDA evals in74.87 s) with Slot-1 Platt calibration (T = 1.67,b = -0.795,conformal_q_hat = 0.5885). - Declarative Gemma 4 ADK Export (
benchmarks/colab_kaggle_l4x4_runner.py --build-submissions): Generates compliantContainer-Aarchives for both tracks (submission_31b_deep.zip83,695 Bandsubmission_e4b_sniper.zip83,384 B,sys.modules["numpy"] = None&torch = Noneparity0.0,0/workspacewrites).
Zero-Shot Blind Verification Scorecard
| Benchmark | Metric | v2.2.0 / Baseline | XLaya v2.3.0 |
|---|---|---|---|
Kaggle 129 (R1 Sidecars) |
E4B R@3 / 31B R@8 / Top-1 ECE |
31.01% / 35.66% / 0.0705 |
65.12% / 78.29% / 0.0568 (80.60% fastapi) |
Kaggle 129 (R2 Pure Stdlib) |
E4B R@3 / 31B R@8 / Top-1 ECE |
— | 62.79% / 78.29% / 0.0641 (98.49% Stage-A tok compression) |
SWE-bench_Verified (n = 94) |
E4B R@3 / 31B R@8 / Top-1 ECE |
24.47% / 26.60% / 0.2910 |
64.89% / 79.79% / 0.0155 (365.4 ms sympy) |
nvidia/When2Call Val (n = 1,616) |
BinAcc (ToolRecall / Abstain) |
62.75% (0% / 100%) |
81.37% (81.40% / 81.36%, ECE = 0.0279) |
nvidia/When2Call Test MCQ (n = 3,652) |
BinAcc (ToolRecall / Abstain) |
64.54% (ECE = 0.3462) |
71.69% (69.11% / 73.10%, ECE = 0.0777, 1.11 ms/row) |
nvidia/When2Call LLM-Judge (n = 300) |
BinAcc (CannotAns / ReqInfo) |
66.67% (ECE = 0.3187) |
68.67% (79.00% / 59.00%, ECE = 0.1210) |
Cross-Domain (xerj#1108 / PLE-29) |
Blind Acc / Wrong-Conf(p≥0.80) |
58.62% nano (PLE) |
100.0% (48/48) & 0.00% / 86.21% (25/29) |
Packaging & Verification Checks
python -B skills/xlaya/scripts/laya_triage.py --self-test→12/12checks passed (version: "2.3.0")python -B -m pytest -q tests/test_v23_sieve_and_adk.py→5/5passedxlaya_mcp_server.py --self-check→{"ok": true, "server": "xlaya-mcp", "version": "2.3.0"}
v2.1.0 — Pure-Local Ensemble Jury, Multi-View Conformal Debiasing, 4-Facet Thorough Scan & Persistent TTT Policy
XLaya v2.1.0 — 100% Pure-Local High-Accuracy Edition
- Stage 1 Dual-Encoder Ensemble Jury (
--laya-model typed|ensemble): Blends0.45 * laya-english + 0.55 * laya-typed-decisionswith automaticdisagreement >= 0.15Stage 2 escalation and continuousmodel_agreement = round(1.0 - disagreement, 4). - Stage 2 Multi-View Position-Debiased Conformal Consensus (
--vega-views 1..3): Evaluates up to 3 cyclic/reversed option permutations, un-permutes posteriors, and intersects/unions conformalprediction_setto eliminate option-order bias. - Persistent Zero-Backprop
TTTHead& Split-Conformal Policy (xlaya_fit/--fit-csv/--policy-path): Fits and persists calibratedTTTHead+ temperatureT+ per-classq_hatpolicies (~/.cache/xlaya/policies/<name>.json,90.0% = 18/20holdout accuracy in0.79 s). - Decomposed 4-Facet Thorough Rubric & Document-Level Aggregation (
--thorough): Scores chunks acrossrelevance(0.35),technical_specificity(0.25),actionability(0.20), andssot_consistency(0.20), plus document-levelfile_summaries. - Explicit Out-of-Distribution (
outbound) Guard & Local Multimodal (image/audio/video): Surfaces normalizedoutbound+ood_scoreon every Stage 2 answer and supports 100% localimage,audio, andvideodecision routing. - 6-Tool Resident JSON-RPC 2.0
stdioMCP Server (xlaya_mcp_server.py): Addsxlaya_fitalongside upgradedxlaya_triage,xlaya_predict,xlaya_status,xlaya_preload, andxlaya_unload.
v2.0.0 — 3-Stage System 1 to System 1.5 Semantic Decision & Conformal Triage Engine
XLaya v2.0.0 (iscvlab/xlaya)
- 3-Stage System 1 -> System 1.5 Cascade (
output_tokens: 0): Stage 0 pure-Python 4-backtick fence-safe multi-format chunker + BM25/code/Thai-3gram sieve (6,841.3 chunks/s,-99.35%Pass-B token compression) -> Stage 1laya 0.4.31+(ModernBERT-large&mmBERT+ Head-Language Guard + 3-Question Joint NLI Contradiction Triad +predict_tournamenton>20labels) -> Stage 2vegaml 0.16.1+vega-Uconformal verifier (up to73,728tokens). - Resident JSON-RPC 2.0
stdioMCP Server (xlaya_mcp_server.py):29.99 mswarm single-state inference,662.49 mswarm 32-chunk Triad (112.5xfaster than v1.0 cold CLI), andLAYA_IDLE_TTL_SEC=1800idle VRAM reclamation. - Empirical 4-Axis RTX 5060 Ti (16 GB VRAM) Benchmarks:
74.51 s -> 9.41 s(7.92x) cold CLI boot;859.1 q/s(5.9x-6.5x)mmBERT&837.9 q/s(8.9x) mixed EN/HI batch routing (0 / 432flips);60.0% -> 90.0%(18/20) 40-row fittedvega-U 0.8Brouting accuracy in7.58 s;8 / 8 (100%)Joint NLI Triad precision; and24.49% -> 0.00%(0/30standalone,0/36ensemble) wrong-and-confident fires onxerj-org/xerj#1108.
