Multimodal models with leading performance.
AI & ML interests
Large Language Models
Recent Activity
View all activity
Papers
Hybrid Linear Attention Done Right: Efficient Distillation and Effective Architectures for Extremely Long Contexts
InfLLM-V2: Dense-Sparse Switchable Attention for Seamless Short-to-Long Adaptation
The first On-device Agent Model Series: Deep Search Exploration and Deep Report Generation.
MiniCPM4: Ultra-Efficient LLMs on End Devices
-
MiniCPM4: Ultra-Efficient LLMs on End Devices
Paper β’ 2506.07900 β’ Published β’ 93 -
openbmb/MiniCPM4.1-8B
Text Generation β’ 8B β’ Updated β’ 15.9k β’ 383 -
openbmb/MiniCPM4.1-8B-GGUF
Text Generation β’ 8B β’ Updated β’ 264 β’ 14 -
openbmb/MiniCPM4-8B
Text Generation β’ 8B β’ Updated β’ 1.53k β’ 281
Extrapolating RLVR to General Domains without Verifiers
-
RLPR: Extrapolating RLVR to General Domains without Verifiers
Paper β’ 2506.18254 β’ Published β’ 32 -
openbmb/RLPR-Train-Dataset
Viewer β’ Updated β’ 77.7k β’ 54 β’ 27 -
openbmb/RLPR-Qwen2.5-7B-Base
Text Generation β’ 8B β’ Updated β’ 41 β’ 8 -
openbmb/RLPR-Gemma2-2B-it
Text Generation β’ 3B β’ Updated β’ 37 β’ 3
UltraLM, UltraRM and UltraCM.
Advancing LLM Reasoning Generalists with Preference Trees
Embedding, re-ranking, generation -- the cornerstone of RAG.
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
-
HuggingFaceH4/zephyr-7b-beta
Text Generation β’ 7B β’ Updated β’ 66.2k β’ β’ 1.83k -
HuggingFaceH4/zephyr-7b-gemma-v0.1
Text Generation β’ 9B β’ Updated β’ 295 β’ 124 -
allenai/tulu-2-dpo-70b
Text Generation β’ 69B β’ Updated β’ 2.17k β’ 158 -
allenai/tulu-2-dpo-13b
Text Generation β’ 13B β’ Updated β’ 1.88k β’ β’ 21
CPM-Bee series models.
Parsing-free RAG supported by VLMs
-
VisRAG 2.0: Evidence-Guided Multi-Image Reasoning in Visual Retrieval-Augmented Generation
Paper β’ 2510.09733 β’ Published β’ 5 -
VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents
Paper β’ 2410.10594 β’ Published β’ 29 -
openbmb/EVisRAG-7B
8B β’ Updated β’ 463 β’ 4 -
openbmb/EVisRAG-3B
4B β’ Updated β’ 99 β’ 1
Multimodal models with leading performance.
NOSA: Native and Offloadable Sparse Attention
The first On-device Agent Model Series: Deep Search Exploration and Deep Report Generation.
MiniCPM4: Ultra-Efficient LLMs on End Devices
-
MiniCPM4: Ultra-Efficient LLMs on End Devices
Paper β’ 2506.07900 β’ Published β’ 93 -
openbmb/MiniCPM4.1-8B
Text Generation β’ 8B β’ Updated β’ 15.9k β’ 383 -
openbmb/MiniCPM4.1-8B-GGUF
Text Generation β’ 8B β’ Updated β’ 264 β’ 14 -
openbmb/MiniCPM4-8B
Text Generation β’ 8B β’ Updated β’ 1.53k β’ 281
The MiniCPM family of LLMs and VLLMs.
Extrapolating RLVR to General Domains without Verifiers
-
RLPR: Extrapolating RLVR to General Domains without Verifiers
Paper β’ 2506.18254 β’ Published β’ 32 -
openbmb/RLPR-Train-Dataset
Viewer β’ Updated β’ 77.7k β’ 54 β’ 27 -
openbmb/RLPR-Qwen2.5-7B-Base
Text Generation β’ 8B β’ Updated β’ 41 β’ 8 -
openbmb/RLPR-Gemma2-2B-it
Text Generation β’ 3B β’ Updated β’ 37 β’ 3
The collection of open-source models that adopt Ultra Series datasets for training
-
HuggingFaceH4/zephyr-7b-beta
Text Generation β’ 7B β’ Updated β’ 66.2k β’ β’ 1.83k -
HuggingFaceH4/zephyr-7b-gemma-v0.1
Text Generation β’ 9B β’ Updated β’ 295 β’ 124 -
allenai/tulu-2-dpo-70b
Text Generation β’ 69B β’ Updated β’ 2.17k β’ 158 -
allenai/tulu-2-dpo-13b
Text Generation β’ 13B β’ Updated β’ 1.88k β’ β’ 21
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 β’ 2510.09733 β’ Published β’ 5 -
VisRAG: Vision-based Retrieval-augmented Generation on Multi-modality Documents
Paper β’ 2410.10594 β’ Published β’ 29 -
openbmb/EVisRAG-7B
8B β’ Updated β’ 463 β’ 4 -
openbmb/EVisRAG-3B
4B β’ Updated β’ 99 β’ 1
Embedding, re-ranking, generation -- the cornerstone of RAG.