OpenOneRec
An Open Foundation Model and Benchmark to Accelerate Generative Recommendation
📖 Introduction
OpenOneRec is an open-source framework designed to bridge the gap between traditional recommendation systems and Large Language Models (LLMs). While Generative Recommendation has shown promise, existing models often struggle with isolated data silos and a lack of reasoning capabilities.
To address this, we introduce a unified framework that comprises:
- RecIF-Bench: The first holistic Recommendation Instruction-Following Benchmark, containing 100M interactions from 200k users across heterogeneous domains (Short Video, Ads, Goods).
- OpenOneRec-Foundation Models: A family of models (1.7B & 8B) built on the Qwen backbone. These models are trained on hundreds of billions of tokens, integrating collaborative signals with general semantics.
- Full-Stack Pipeline: We open-source our comprehensive training pipeline, including data processing, co-pretraining, and post-training, to ensure full reproducibility and facilitate scaling law research in recommendation.
🔥 News
- [2025.xx.xx] 🎉 OpenOneRec-Foundation models (1.7B, 8B) are now available on Hugging Face!
- [2025.xx.xx] 📑 The technical report [OpenOneRec Technical Report] has been released.
- [2025.xx.xx] 🚀 RecIF-Bench dataset and evaluation scripts are open-sourced.
📊 RecIF-Bench
We propose RecIF-Bench to rigorously assess the synergy between instruction following and domain-specific recommendation. It organizes 9 distinct tasks into a four-layer capability hierarchy:
- Layer 0: Semantic Alignment (Item Understanding)
- Layer 1: Fundamental Prediction (Short Video Rec, Ad Rec, Goods Rec, Label Prediction)
- Layer 2: Instruction Following (Interactive Rec, Label-Conditional Rec)
- Layer 3: Reasoning (User Summarization, Recommendation Explanation)
The benchmark aggregates data from three domains: Short Video (Content), Ads (Commercial), and Goods (E-commerce).
🤖 Model Zoo
The OpenOneRec-Foundation series is built upon the Qwen architecture, enhanced with Itemic Tokens for modality alignment and trained via a multi-stage protocol.
| Model | Backbone | Parameters | Description | Link |
|---|---|---|---|---|
| OneRec-Open-1.7B | Qwen3-1.7B | 1.7B | Standard version trained on open-source data (~100B tokens) | HuggingFace |
| OneRec-Open-8B | Qwen3-8B | 8B | Standard version trained on open-source data (~100B tokens) | HuggingFace |
| OneRec-Pro-1.7B | Qwen3-1.7B | 1.7B | Enhanced version with proprietary tokens | Coming Soon |
| OneRec-Pro-8B | Qwen3-8B | 8B | Enhanced version with proprietary tokens | Coming Soon |
🏗️ Method & Architecture
OpenOneRec reframes recommendation as a general-purpose sequence modeling paradigm.
1. Items as Tokens
To bridge the modality gap, we treat items as a distinct modality using Itemic Tokens derived from hierarchical vector quantization. This allows the LLM to process interaction history as a cohesive context sequence.
2. Training Pipeline
Our framework utilizes the following recipe:
- Pre-Training: Integrates collaborative signals via Itemic-Text Alignment and mixed-domain Co-Pretraining.
- Post-Training:
- Stage 1: Cold-Start SFT for basic instruction following.
- Stage 2: Alternating On-Policy Distillation & SFT to balance general reasoning and recommendation performance.
- Stage 3: Recommendation-oriented Reinforcement Learning (Rec-RL).
Figure: The Overall Framework of OpenOneRec.
📈 Performance
Results on RecIF-Bench
OpenOneRec-Foundation achieves State-of-the-Art (SOTA) results across RecIF-Bench tasks, significantly outperforming baselines like LC-Rec and TIGER.
| Task | Metric | TIGER | LC-Rec | OneRec-Pro-8B |
|---|---|---|---|---|
| UserDoc | Recall@32 | 0.0132 | 0.0180 | 0.0369 |
| Ads Rec | Recall@32 | 0.0581 | 0.0723 | 0.0964 |
| Goods Rec | Recall@32 | 0.0283 | 0.0416 | 0.0538 |
Cross-Domain Transferability
On the Amazon Benchmark (10 datasets), OpenOneRec demonstrates exceptional zero-shot/few-shot transfer capabilities, achieving an average 26.8% improvement in Recall@10 over the second-best method.
🚀 Quick Start
Code release and detailed usage instructions are coming soon.
Currently, you can load our models using transformers:
from transformers import AutoModelForCausalLM, AutoTokenizer
model_id = "Onerec/OneRec-Open-8B"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
# Example inference code will be updated here
📜 Citation
If you find our work helpful, please cite our technical report:
@article{openonerec2025,
title={An Open Foundation Model and Benchmark to Accelerate Generative Recommendation},
author={OneRec Team},
journal={arXiv preprint},
year={2025}
}
🛡️ License
The code in this repository is licensed under the Apache 2.0 License. The model weights are subject to their specific license agreements.