TabPFN-2.5 is a tabular foundation model based on the Transformer architecture. It utilizes context learning technology and can solve tabular prediction problems in a single forward propagation, providing efficient regression and classification solutions for structured tabular data.

Kimi Linear is a hybrid linear attention architecture that outperforms traditional full attention methods in various scenarios, including short and long contexts, as well as reinforcement learning extension mechanisms. It can effectively solve the problem of low efficiency of traditional attention mechanisms in long context tasks and bring more efficient solutions to fields such as natural language processing.

Natural Language ProcessingTransformers

This is an image text recognition model based on DeepSeek-OCR, which specifically solves the compatibility issues in the latest version of the transformers library, enabling the model to run smoothly in the latest versions such as transformers v4.57.1.

MultimodalTransformersEnglish

This is a repaired image text-to-text model that solves the problem of inference failure of the original model in the latest Transformers version. This model is specifically designed for image recognition and text generation tasks and supports multimodal input processing.

MultimodalTransformersEnglish

Fathom-Search-4B GGUF is a powerful tool designed for text generation tasks. It is generated based on specific base models and technologies and performs excellently in search-intensive benchmark tests. It can effectively solve the problems of long-cycle information retrieval and synthesis.

Natural Language ProcessingTransformers

BR-RM is an innovative two-round reasoning reward model that solves the 'judgment diffusion' problem in traditional reward models through adaptive branching and branch-based reflection mechanisms, achieving industry-leading performance in multiple reward modeling benchmark tests.

Natural Language ProcessingTransformersEnglish

Rapido NER is a powerful multilingual named entity recognition and entity embedding model that supports 55 languages. It integrates an encoder, attention-based mention pooling, a type projection layer, and conditional random field decoding, specifically designed to solve common problems in the field of named entity recognition.

Natural Language ProcessingTransformersMultiple Languages

SRPO is a human preference alignment method for diffusion models. Through Direct-Align technology and semantic relative preference optimization, it significantly improves the realism and aesthetic quality of the FLUX.1-dev model, and solves the problems of high computational cost in multi-step denoising and dependence on offline reward fine-tuning.

Computer VisionDiffusers

This is the GGUF quantized version of the Qwen3-Code-Reasoning-4B model, specifically optimized for competition programming and code reasoning tasks. The model is trained on a high-quality code reasoning dataset, enhancing its ability to solve complex programming problems through detailed reasoning.

Natural Language ProcessingGgufEnglish

Datarus-R1-14B-Preview is an open-source language model based on 14 billion parameters, fine-tuned on the basis of Qwen2.5-14B-Instruct. This model is specifically designed as a virtual data analyst and a solver for graduate-level problems, capable of learning from complete analysis trajectories, including reasoning steps, code execution, error tracking, and final conclusions.

Natural Language ProcessingTransformersEnglish

Kimina-Prover-Distill-0.6B is a theorem proving model developed by Project Numina and the Kimi team, focusing on the ability to solve competition-style problems in Lean 4. It is a distilled version of the Kimina-Prover-72B model, achieving an accuracy of 68.85% on MiniF2F-test.

Natural Language ProcessingTransformersEnglish

The GPT-OSS series is an open-weight model launched by OpenAI, designed for powerful inference capabilities, agent tasks, and diverse developer use cases. This model can effectively solve complex problems, provide a flexible and efficient usage experience, and support the Apache 2.0 open-source license.

Natural Language ProcessingTransformers

AceMath-7B-Instruct is a cutting-edge model specifically designed for mathematical reasoning. It is developed based on Qwen2.5-Math-7B-Base and is proficient in using thought chain (CoT) reasoning to solve English mathematical problems.

Natural Language ProcessingGgufEnglish

A theorem proving model developed by the Project Numina and Kimi teams, focusing on the ability to solve competition-style problems in Lean 4.

Scientific ComputingTransformersEnglish

SWE-agent-LM-7B is an open-source language model specifically designed for the software engineering field. It is fine-tuned based on Qwen2.5-Coder-7B-Instruct. This model can effectively solve software engineering-related problems and provide intelligent support for developers, which has important practical value.

Natural Language ProcessingTransformersEnglish

A theorem proving model developed by the Numina project and the Kimi team, focusing on improving the ability to solve competition-style problems in Lean 4.

Natural Language ProcessingTransformersEnglish

Kimina-Prover-Distill-8B is a theorem proving model developed by the Project Numina and Kimi teams, focusing on the ability to solve competition-style problems in Lean 4.

Natural Language ProcessingTransformersEnglish

An 11B parameter coding AI model based on the fusion of Qwen2.5-Coder-7B and OlympicCoder-7B, which can quickly solve programming problems and generate high-quality code

Natural Language ProcessingGgufMultiple Languages

A vision-language model based on the OpenCLIP-ViT-B-16 architecture. By introducing test-time registers to optimize the internal representation, it solves the problem of feature map artifacts.

MultimodalTransformers

The server-side seal text detection model of PP-OCRv4, with high accuracy, suitable for server deployment, and can effectively solve the problem of seal text detection.