MiniOneRec: An Open-Source Framework for Scaling Generative Recommendation

2025-10-28

AI Summary

This work investigates whether large language models (LLMs) can exhibit scaling-law-like performance gains in recommendation systems—overcoming the saturation bottleneck of traditional embedding-table-based methods in high-dimensional spaces. We propose the first fully open-source generative recommendation framework: it replaces fixed embedding tables with semantically meaningful ID sequences generated by a residual-quantized VAE, performs supervised fine-tuning and recommendation-oriented reinforcement learning atop Qwen-series LLMs, and introduces a lightweight post-training procedure for end-to-end semantic alignment and constrained decoding. Evaluated on Amazon Reviews, our approach demonstrates consistent improvements in ranking accuracy and candidate diversity with increasing model scale—significantly outperforming baselines while achieving higher parameter efficiency. The core contribution is the empirical validation of scalability within the generative paradigm for recommendation, coupled with an end-to-end, reproducible, open-source implementation pipeline.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

The recent success of large language models (LLMs) has renewed interest in whether recommender systems can achieve similar scaling benefits. Conventional recommenders, dominated by massive embedding tables, tend to plateau as embedding dimensions grow. In contrast, the emerging generative paradigm replaces embeddings with compact Semantic ID (SID) sequences produced by autoregressive Transformers. Yet most industrial deployments remain proprietary, leaving two fundamental questions open: (1) Do the expected scaling laws hold on public benchmarks? (2) What is the minimal post-training recipe that enables competitive performance? We present MiniOneRec, to the best of our knowledge, the first fully open-source generative recommendation framework, which provides an end-to-end workflow spanning SID construction, supervised fine-tuning, and recommendation-oriented reinforcement learning. We generate SIDs via a Residual Quantized VAE and post-train Qwen backbones ranging from 0.5B to 7B parameters on the Amazon Review dataset. Our experiments reveal a consistent downward trend in both training and evaluation losses with increasing model size, validating the parameter efficiency of the generative approach. To further enhance performance, we propose a lightweight yet effective post-training pipeline that (1) enforces full-process SID alignment and (2) applies reinforcement learning with constrained decoding and hybrid rewards. Together, these techniques yield significant improvements in both ranking accuracy and candidate diversity.

Problem

• Validating scaling laws for generative recommender systems
• Developing minimal post-training for competitive recommendation performance
• Creating open-source framework for end-to-end generative recommendation workflow

Innovation

• Generative framework replaces embeddings with Semantic IDs
• Residual Quantized VAE constructs compact item representations
• Reinforcement learning enhances ranking accuracy and diversity