darrencxl0301/StageRAG
A lightweight RAG framework offering switchable 3-step (Speed) and 4-step (Precision) pipelines built on quantized Llama 3.2 1B/3B models.
Not currently ranked — collecting fresh signals.
star history
StageRAG is a production-ready framework for building hallucination-resistant RAG applications. It provides dual-mode pipelines that let users choose between speed (3-step, ~3-5s) and precision (4-step, ~6-12s) based on their needs. The framework integrates with knowledge bases via JSONL files, automatically builds vector indices for retrieval, and includes multi-component confidence scoring to detect uncertainty and reduce hallucinations. It runs on quantized Llama 3.2 1B and 3B models requiring 5-10GB GPU memory.
Frequently asked
- What is darrencxl0301/StageRAG?
- A lightweight RAG framework offering switchable 3-step (Speed) and 4-step (Precision) pipelines built on quantized Llama 3.2 1B/3B models.
- Is StageRAG open source?
- Yes — darrencxl0301/StageRAG is open source, released under the MIT license.
- What language is StageRAG written in?
- darrencxl0301/StageRAG is primarily written in Python.
- How popular is StageRAG?
- darrencxl0301/StageRAG has 1k stars on GitHub.
- Where can I find StageRAG?
- darrencxl0301/StageRAG is on GitHub at https://github.com/darrencxl0301/StageRAG.