rayliuca/T-Ragx

Enhancing Translation with RAG-Powered Large Language Models

51
/ 100
Established

Combines QLoRA fine-tuned models with Elasticsearch-backed retrieval to inject domain-specific translation memories and glossaries into LLM context, enabling document-level awareness through preceding text. Supports multiple model backends (Hugging Face, Ollama, OpenAI, llama-cpp-python) and demonstrates 45% WMT23 score improvements over base models by leveraging in-task RAG strategies. Prioritizes local execution for privacy while maintaining recall on terminology and character consistency across translations.

Available on PyPI.

Maintenance 6 / 25
Adoption 9 / 25
Maturity 25 / 25
Community 11 / 25

How are scores calculated?

Stars

95

Forks

8

Language

Python

License

MIT

Last pushed

Dec 29, 2025

Commits (30d)

0

Dependencies

5

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/rag/rayliuca/T-Ragx"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.