SOTA on-device LLMs, small yet powerful.
AI & ML interests
Large Language Models
Recent Activity
Papers
SimpleMemVLA: A Simple but Effective Native-Video Memory for Vision-Language-Action Models
MathForm: Scaling Mathematical Autoformalization with Knowledge Retrieval and Verification-Guided Refinement
Full-pipeline ternary quantized model trained on CANN.
Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
Multimodal models with leading performance.
-
MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction
Paper • 2604.27393 • Published • 85 -
MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe
Paper • 2509.18154 • Published • 63 -
MiniCPM-V: A GPT-4V Level MLLM on Your Phone
Paper • 2408.01800 • Published • 96 -
openbmb/MiniCPM-V-4.6
Image-Text-to-Text • 1B • Updated • 507k • 1.21k
MiniCPM4: Ultra-Efficient LLMs on End Devices
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 • 143 • 6 -
openbmb/EVisRAG-3B
4B • Updated • 138 • 2
A Smarter and Faster On-Device AI Brain for Robots
A Pocket-Sized MLLM for Ultra-Efficient Image and Video Understanding on Your Phone
Embedding, re-ranking, generation -- the cornerstone of RAG.
Ultra Scale, Ultra Quality, Ultra Coverage
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
SOTA on-device LLMs, small yet powerful.
A Smarter and Faster On-Device AI Brain for Robots
Full-pipeline ternary quantized model trained on CANN.
A Pocket-Sized MLLM for Ultra-Efficient Image and Video Understanding on Your Phone
Tokenizer-Free TTS for Multilingual Speech Generation, Creative Voice Design, and True-to-Life Cloning
Embedding, re-ranking, generation -- the cornerstone of RAG.
Multimodal models with leading performance.
-
MiniCPM-o 4.5: Towards Real-Time Full-Duplex Omni-Modal Interaction
Paper • 2604.27393 • Published • 85 -
MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe
Paper • 2509.18154 • Published • 63 -
MiniCPM-V: A GPT-4V Level MLLM on Your Phone
Paper • 2408.01800 • Published • 96 -
openbmb/MiniCPM-V-4.6
Image-Text-to-Text • 1B • Updated • 507k • 1.21k
Ultra Scale, Ultra Quality, Ultra Coverage
MiniCPM4: Ultra-Efficient LLMs on End Devices
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 • 143 • 6 -
openbmb/EVisRAG-3B
4B • Updated • 138 • 2