AI & ML interests
Large Language Models
Recent Activity
Papers
MiniCPM-SALA: Hybridizing Sparse and Linear Attention for Efficient Long-Context Modeling
Data Science and Technology Towards AGI Part I: Tiered Data Management
MiniCPM4: Ultra-Efficient LLMs on End Devices
FactNet: A Billion-Scale Knowledge Graph for Multilingual Factual Grounding
The first On-device Agent Model Series: Deep Search Exploration and Deep Report Generation.
Extrapolating RLVR to General Domains without Verifiers
UltraLM, UltraRM and UltraCM.
Advancing LLM Reasoning Generalists with Preference Trees
Embedding, re-ranking, generation -- the cornerstone of RAG.
Multimodal models with leading performance.
Ultra Scale, Ultra Quality, Ultra Coverage
NOSA: Native and Offloadable Sparse Attention
The MiniCPM family of LLMs and VLLMs.
The collection of open-source models that adopt Ultra Series datasets for training
CPM-Bee series models.
Parsing-free RAG supported by VLMs
-
VisRAG 2.0: Evidence-Guided Multi-Image Reasoning in Visual Retrieval-Augmented Generation
Paper • 2510.09733 • Published • 5 -
VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents
Paper • 2410.10594 • Published • 29 -
openbmb/EVisRAG-7B
8B • Updated • 52 • 5 -
openbmb/EVisRAG-3B
4B • Updated • 10.4k • 2
Multimodal models with leading performance.
MiniCPM4: Ultra-Efficient LLMs on End Devices
Ultra Scale, Ultra Quality, Ultra Coverage
FactNet: A Billion-Scale Knowledge Graph for Multilingual Factual Grounding
NOSA: Native and Offloadable Sparse Attention
The first On-device Agent Model Series: Deep Search Exploration and Deep Report Generation.
The MiniCPM family of LLMs and VLLMs.
Extrapolating RLVR to General Domains without Verifiers
The collection of open-source models that adopt Ultra Series datasets for training
UltraLM, UltraRM and UltraCM.
CPM-Bee series models.
Advancing LLM Reasoning Generalists with Preference Trees
Parsing-free RAG supported by VLMs
-
VisRAG 2.0: Evidence-Guided Multi-Image Reasoning in Visual Retrieval-Augmented Generation
Paper • 2510.09733 • Published • 5 -
VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents
Paper • 2410.10594 • Published • 29 -
openbmb/EVisRAG-7B
8B • Updated • 52 • 5 -
openbmb/EVisRAG-3B
4B • Updated • 10.4k • 2
Embedding, re-ranking, generation -- the cornerstone of RAG.