evalution

Evaluation tools for Retrieval-augmented Generation (RAG) methods.

Evidence-grounded multi-source knowledge graph construction with duplicate-resistant identity resolution, provenance, validation, and reproducible benchmarks.

The official repository for the dataset FactualBench, which is introduced in paper "Exploring the Generalizability of Factual Hallucination Mitigation via Enhancing Precise Knowledge Utilization".

A Python toolkit for the robust evaluation and analysis of seizure detection models.

A Flask-Celery integration template for distributed task processing with progress tracking and use MMMLU to evaluation example.

Production-oriented RAG pipeline for internal documents using hybrid dense + BM25 search, RRF fusion, cross-encoder reranking, confidence gating, and verified citations. Includes FastAPI APIs, Qdrant storage, evaluation benchmarks, testing, Docker deployment, and a monitoring dashboard.

OpenCompass is an LLM evaluation platform, supporting a wide range of models (Llama3, Mistral, InternLM2,GPT-4,LLaMa2, Qwen,GLM, Claude, etc) over 100+ datasets.

LLM mock-interview agent: state-machine interview flow, structured scoring stored in SQLite, Discord bot and Streamlit dashboard. Includes an evaluation showing the judge over-scores fluent but wrong answers, and a fact-check fix.

Read-only GitHub MCP server + a tool-selection benchmark: 63% → 95% on a local 7B model, with an ablation of where the gain came from.