ChainCheck — Chain-Sensitive Evidence Judges - a ThakiCloud Collection

ThakiCloud 's Collections

Thaki-HW — Hardware Design LLM (Research Preview)

EmbeddingGemma 2 Edge — Quantization

RAG-Gate — Answer / Retrieve / Stop Gates for RAG

ChainCheck — Chain-Sensitive Evidence Judges

PACE — Enterprise Policy Decisions: Learned or Reasoned

Merchant Desk — Korean Small-Business Customer Desk

Qwen-Image-2.1 — Few-Step Distillation

K-Decision — Korean typed-decision adapters (weights only)

K-MT — Korean-Specialized Translation Without Forgetting

Satoori-KO — Korean Regional Dialect 27B

Evals & Reproducibility

Language Confusion Suppression — Vocabulary Pruning

Vision-Language — Serving Quantization

Qwen3-30B-A3B — Serving Quantization

Qwen3-Coder-30B-A3B — Serving Quantization

Qwen3.8-27B — Serving Quantization

SkillRet — Agent Skill Retrieval

Human-KO — Korean Style-Aligned 27B

ChainCheck — Chain-Sensitive Evidence Judges

updated 6 days ago

Evidence-sufficiency judges for RAG (4B/9B/27B) tuned with ChainCheck controls, plus the kit. Gains are on real entities, not synthetic ones.

ThakiCloud/ChainCheck-Judge-Qwen3.8-27B

Text Generation • 28B • Updated 6 days ago • 545 • 2

Note 27B judge, merged bf16 weights. 2Wiki chain selectivity +0.129 → +0.461 vs. zero-shot; Σ lower bound > 0 on 6/6 slices
ThakiCloud/ChainCheck-Judge-Qwen3.5-9B

Text Generation • 10B • Updated 6 days ago • 358

Note 9B judge, merged bf16 weights. 2Wiki chain selectivity +0.078 → +0.386 vs. zero-shot; Σ lower bound > 0 on 6/6 slices
ThakiCloud/ChainCheck-Judge-Qwen3.5-4B

Text Generation • 5B • Updated 6 days ago • 352

Note 4B judge, merged bf16 weights. 2Wiki chain selectivity +0.136 → +0.376 vs. zero-shot; Σ lower bound > 0 on 5/6 slices
ThakiCloud/ChainCheck

Viewer • Updated 6 days ago • 3.79k • 68

Note Benchmark kit — A/B/D cells on MuSiQue and 2WikiMultiHopQA (3,786 items) + numpy-only evaluator