This is the quantized version of Qwen3-VL-32B-Thinking-heretic, using a new ablation method for quantization. Compared with traditional methods, it causes less damage to the base model. It offers multiple quantization precision versions and is suitable for running on devices with 24GB of VRAM, supporting image-to-text tasks.

MultimodalTransformersEnglish

Gheya-1 is a new-generation basic language model in the LES-IA-ETOILES ecosystem, with 202 million parameters. It is an upgraded version of the old Small-lamina series. This model is designed specifically for professional fine-tuning and has targeted training in the fields of artificial intelligence, professional language models, and biology.

Natural Language ProcessingTransformersFrench

Nanbeige 4-3B-Thinking-2511 is the latest enhanced version of the Nanbeige series. Through advanced distillation technology and reinforcement learning optimization, it achieves powerful reasoning ability within a compact 3B parameter scale. This model has achieved state-of-the-art (SOTA) results among models with less than 32B parameters in benchmark tests such as Arena-Hard-V2 and BFCL-V4.

Natural Language ProcessingTransformersMultiple Languages

WavJEPA is an audio foundation model based on the waveform joint embedding prediction architecture. It uses advanced semantic representation learning to address the deficiencies in speech unit or token-level representation learning. It significantly outperforms the state-of-the-art time-domain audio foundation models in many downstream benchmark tasks while requiring significantly less computational resources.

Audio ProcessingTransformers

This is a sequence-to-sequence model specifically designed to compress short user prompts (≤64 tokens) and can achieve low-latency operation of less than 100 milliseconds on modern GPUs. It serves as a lightweight preprocessing stage to support high-capacity large language models.

Natural Language ProcessingTransformers

LightOnOCR-1B-1025 is a compact end-to-end vision-language model specifically designed for optical character recognition and document understanding. It achieves state-of-the-art accuracy at the same weight level, while being faster and less costly than large general-purpose vision-language models.

MultimodalSafetensorsMultiple Languages

MobileLLM-R1 is a series of efficient inference models focused on mathematics, programming, and scientific problems, achieving excellent performance with less training data and providing a complete training recipe and data source.

Natural Language ProcessingTransformersEnglish

VaultGemma is a lightweight language model pre-trained by Google based on differential privacy technology, with powerful privacy protection capabilities. It uses differential privacy stochastic gradient descent (DP-SGD) for pre-training, providing mathematical privacy guarantees for the training data. It has less than 1 billion parameters and is suitable for a variety of natural language processing tasks.

Natural Language ProcessingTransformers

This model is fine-tuned based on the DINOv2 architecture for disease classification of skin lesion images

Computer VisionPytorchEnglish

This is a fine-tuned vision transformer model specifically designed for classifying 12 types of skin lesions. The model is based on the pre-trained ViT architecture and has been fine-tuned on approximately 70k skin lesion images. It can accurately identify various skin diseases, including melanoma and basal cell carcinoma.

Computer VisionSafetensors

Illustrious XL is a text-to-image model based on Stable Diffusion XL, specializing in generating photorealistic images, with particular expertise in portraying female figures and lesbian themes.

Computer VisionDiffusersEnglish

A 2.5D style text-to-image model based on Stable Diffusion XL, supporting photorealistic and artistic style generation, particularly skilled in female imagery and lesbian-themed creations.

Computer VisionDiffusersEnglish

This is a pruned version of the BAAI/bge-m3 model optimized for Arabic, retaining approximately 98% of the original model's quality while using less memory.

Natural Language ProcessingSafetensorsMultiple Languages

A 3D deep learning model based on multi-scale supervised UNet for small vessel segmentation in 7T MRA-ToF volumetric data, specifically designed for research on cerebral microvascular lesions.

Computer VisionSafetensors

CodeGeeX4-ALL-9B is a multilingual code generation model continuously trained based on GLM-4-9B. It is the open-source version of the CodeGeeX4 model series. This model performs the best among code generation models with less than 10 billion parameters. It supports comprehensive functions such as code completion, code interpreter, web search, function call, and repository-level code Q&A, achieving the best balance between inference speed and model performance.

Natural Language ProcessingGgufMultiple Languages

CodeGeeX4 is a multilingual code generation model continuously trained based on GLM-4-9B. It is currently the most powerful code generation model with less than 10 billion parameters. It supports comprehensive functions such as code completion, code generation, code explanation, web search, function call, and repository-level code Q&A, achieving the best balance between inference speed and model performance.

Natural Language ProcessingTransformersMultiple Languages

WhyLesionCLIP is a fine-tuned model based on OpenCLIP (ViT-L/14) for aligning skin lesion images with text descriptions, trained on the ISIC dataset, supporting zero-shot skin lesion classification.

MultimodalOpenclip

Vision Transformer (ViT)-based skin cancer image classification model capable of identifying 7 types of skin lesions

Computer VisionTransformers

SkinSAM is a skin lesion segmentation model based on a 12-layer ViT-b architecture, fine-tuned on ISIC and PH2 datasets, focusing on precise segmentation of skin lesion images.

Computer VisionTransformersMultiple Languages

QuaLA-MiniLM is a compact language model developed by Intel, integrating knowledge distillation, length-adaptive transformers, and 8-bit quantization technology. It achieves up to 8.8x acceleration on the SQuAD1.1 dataset with less than 1% accuracy loss.

Natural Language ProcessingTransformers