Qwen3 VL 4B Thinking is a multimodal model that supports image-to-text and text-to-text conversion. It has a parameter scale of 4B and can meet various image-text interaction needs.

MultimodalGguf

This repository contains a 4-step distilled model optimized for HunyuanVideo-1.5, which can achieve ultra-fast 4-step inference without using CFG (Classifier-Free Guidance), significantly reducing the generation time while maintaining high-quality video output.

MultimodalDiffusers

Qwen3-VL-4B-Instruct is a 4-billion-parameter visual language model launched by Alibaba. It is developed based on the Qwen3 architecture and supports multimodal understanding and dialogue tasks. This model has powerful image understanding and text generation capabilities, and can handle complex visual language interaction scenarios.

MultimodalTransformers

This project provides a quantized GGUF version of Tencent's Hunyuan Video 1.5 text-to-video model, which supports 480P video generation tasks. It includes both distilled and full models and can be used in conjunction with tools such as ComfyUI-GGUF.

MultimodalGguf

The quantized GGUF version of Tencent's Hunyuan Video 1.5 model, specifically designed for image-to-video and video generation tasks. It offers distilled and standard models with a 480P resolution and supports multiple quantization precisions, including Q4_K_S, Q8_0, and FP16, etc.

MultimodalGguf

UltraFlux is a diffusion transformer based on Flux, specifically designed for native 4K text-to-image generation. Through the collaborative design of data, architecture, and loss, it can maintain consistent image quality under various aspect ratios.

Computer VisionDiffusersEnglish

This project provides the GGUF format file of the Qwen3-4B-Thinking-2507 model. It is a large language model with 4B parameters and chain of thought reasoning ability, supporting multiple quantization versions for easy deployment and operation on various hardware.

Natural Language ProcessingGguf

VibeThinker-1.5B is a 1.5-billion-parameter dense language model launched by Weibo AI. It is fine-tuned based on Qwen2.5-Math-1.5B and is specifically designed for mathematical and algorithmic coding problems. Trained using the 'Spectrum to Signal Principle' framework, it outperforms larger models in multiple math competition tests. The training cost is approximately $7,800, and it supports an output of up to about 40k tokens.

Natural Language ProcessingTransformersMultiple Languages

DeepMath is a mathematical reasoning model with 4 billion parameters, built on Qwen3 - 4B Thinking, combining a fine - tuned large language model and a sandboxed Python executor. It can generate concise Python code snippets for calculation steps, significantly reducing errors and shortening the output length.

Natural Language ProcessingTransformersEnglish

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

The 4-bit quantized version of VibeThinker-1.5B, optimized for Apple chips based on the MLX framework, is a dense language model with 1.5 billion parameters, specifically designed for mathematical reasoning and algorithm coding problems.

Natural Language ProcessingMlxMultiple Languages

This is a large language model with 139B parameters obtained by uniformly pruning 40% of the experts in MiniMax-M2 using the REAP method. It adopts the GLM architecture and Mixture of Experts (MoE) technology, and undergoes various quantization processes through llama.cpp, suitable for text generation tasks.

Natural Language ProcessingGgufEnglish

This is a quantized version of the p-e-w/gpt-oss-20b-heretic model, quantized using the imatrix technology of llama.cpp. The model is a large language model with 20 billion parameters, offering a variety of quantization options ranging from high quality to low quality, with file sizes from 41.86GB to 11.52GB, suitable for different hardware conditions.

Natural Language ProcessingGguf

Wan2.2-TI2V-5B-Turbo-GGUF is an image-to-video model converted from the quanhaol/Wan2.2-TI2V-5B-Turbo base model. It has been optimized to run on a 4GB GPU and has excellent versatility and efficient inference capabilities.

MultimodalGguf

This is a large language model with 30 billion parameters trained on the Claude Sonnet 4.5 reasoning dataset. It is specifically optimized for high-difficulty reasoning tasks and performs excellently in the fields of coding and science.

Natural Language ProcessingTransformersEnglish

This is a distilled model based on the Qwen3 30B A3B model, trained on the Claude Sonnet 4.5 high-reasoning-difficulty dataset. This model is specifically optimized for complex reasoning tasks and performs excellently in the fields of coding and science, with strong logical reasoning ability.

Natural Language ProcessingTransformersEnglish

This is a fine-tuned version of the Helsinki-NLP/opus-mt-en-ber model, specifically designed for English to Atlas Tamazight translation. The model is based on the MarianMT architecture and trained with 486,000 synthetic sentence pairs, supporting translation tasks for Tamazight, Tashelhit, and Central Atlas Tamazight.

Natural Language ProcessingSafetensorsMultiple Languages

This is a mixed quantization version based on the Qwen3 4B Instruct model, using MXFP4 and mixed weight techniques to achieve a smaller file size and higher inference speed while maintaining nearly lossless precision.

Natural Language ProcessingSafetensorsOther

This is a mixed quantization version based on the Qwen3 4B model, using the MXFP4_MOE mixed weight technology. It achieves a smaller file size and higher inference speed while maintaining near-lossless accuracy. By carefully combining MXFP4 and high-precision embedding/output weights, the model reaches an accuracy level close to Q8 quantization and has a throughput at the Q4 - Q6 level.

Natural Language ProcessingSafetensorsOther

An image-to-video model launched by the FastVideo team, belonging to the CausalWan2.2 I2V A14B series. It supports 8-step inference, can adapt to various GPUs from H100 to 4090, and is also available for Mac users.

Computer VisionDiffusers