zjunlp/OneGen
[EMNLP 2024] OneGen: Efficient One-Pass Unified Generation and Retrieval for LLMs.
Integrates token-level role assignment (GEN, CTX, RET) during training to apply cross-entropy loss for generation tokens and contrastive loss for retrieval tokens, enabling joint learning in a single forward pass. Supports multiple downstream tasks (entity linking, single/multi-hop QA) with Llama2-7B checkpoints across HuggingFace, ModelScope, and WiseModel, while leveraging Faiss for efficient vector retrieval at inference time without requiring separate query encoding passes.
147 stars. No commits in the last 6 months.
Stars
147
Forks
15
Language
Python
License
MIT
Category
Last pushed
Nov 13, 2024
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/rag/zjunlp/OneGen"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
mongodb-developer/GenAI-Showcase
GenAI Cookbook
SAP-samples/btp-cap-genai-rag
Explore this repository for GenAI samples on SAP Business Technology Platform (SAP BTP). We...
Denis2054/Building-Business-Ready-Generative-AI-Systems
This GitHub repository contains the complete code for building Business-Ready Generative AI...
deepset-ai/haystack-cookbook
👩🏻🍳 A collection of example notebooks using Haystack
AdilShamim8/GenAI-Roadmap-with-Notes-Using-LangChain
A comprehensive learning path and practical guide for Generative AI development with hands-on...