# Multimodal Understanding

Cosmos-Reason1-7B is a 7B-parameter foundational model released by NVIDIA, specializing in image-to-text tasks.

Vision encoder based on Mistral Small model, supports image-text generation tasks, compatible with llama.cpp framework

Magma-8B is an image-text-to-text conversion model based on the GGUF format, suitable for multimodal task processing.

A vision-language model specifically designed for Thai-English real-world document parsing, based on the Qwen2.5-VL-Instruction framework

A quantized version of the Qwen2.5-VL-72B-Instruct multimodal large language model, supporting image-text-to-text tasks, suitable for various quantization levels from high precision to low memory requirements.

A quantized version of Qwen2.5-VL-7B-Instruct, using llama.cpp for quantization, supporting multimodal tasks such as image-to-text conversion.

A vision-language model fine-tuned on VQA datasets based on the ViLT-B32-MLM model

TEMPURA is a video temporal understanding framework combining causal reasoning with fine-grained temporal segmentation, enhancing video event comprehension through two-stage training

A camera motion analysis model fine-tuned based on Qwen2.5-VL-7B-Instruct, focusing on camera motion classification in videos and video-text retrieval tasks

Gemma 3 is Google's lightweight open model series based on Gemini technology. The 12B version employs Quantization-Aware Training (QAT) technology, supports multimodal input, and features a 128K context window.

Gemma 3 is a lightweight open model series built by Google based on Gemini technology, supporting multimodal input and text output, featuring a 128K large context window and support for 140+ languages.

Gemma 3 is a lightweight open model series from Google, built on the research and technology used to create Gemini models. The 12B version is an instruction-tuned multimodal model supporting text and image inputs to generate text outputs.

A fine-tuned vision-language model based on Salesforce/blip2-opt-2.7b for visual question answering tasks

BLIP is a unified vision-language pretraining framework, excelling in vision-language tasks such as image caption generation

InternVL3-8B-6bit is a vision-language model converted to MLX format, supporting multilingual image-text-to-text tasks.

Gemma 3 12B IT is a large language model developed by Google, supporting multimodal input and long-context processing.

The Gemma 3 4B IT model by Google supports multimodal input and long-context processing, suitable for text generation and image understanding tasks.

This model is a 3-bit quantized version converted from google/gemma-3-27b-it-qat-q4_0-unquantized to the MLX format, suitable for image-to-text tasks.

Gemma 3 27B IT QAT 4bit is an MLX-format model converted from Google's original model, supporting image-to-text tasks.

INT4 quantized version based on the gemma-3-4b-it model, significantly reducing storage and computational resource requirements

Gemma 3 is a lightweight open-source multimodal model series developed by Google, built on Gemini technology, supporting text and image inputs with text outputs. The 1B version has undergone instruction tuning and quantization-aware training (QAT), making it suitable for deployment in resource-constrained environments.

Gemma 3 is Google's lightweight open-source multimodal model series based on Gemini technology, supporting text and image inputs with text outputs. The 12B version undergoes instruction tuning and quantization-aware training (QAT), making it suitable for deployment in resource-limited environments.

Llama 4 Scout is a 17-billion-parameter Mixture of Experts (MoE) model released by Meta, supporting multilingual text and image understanding with a linearized expert module design for PEFT/LoRA compatibility.

Gemma is a family of lightweight, cutting-edge open models introduced by Google, built on the same research and technology as the Gemini models. Supports text and image inputs and generates text outputs.

Gemma 3 is a lightweight, cutting-edge open model series from Google, built on the same technology as Gemini, supporting multimodal input (text/image) and text output. The 27B version significantly reduces memory requirements through quantization-aware training.

Gemma 3 is Google's lightweight open model series based on Gemini technology, supporting multimodal input and text output.

Gemma 3 is a lightweight open-source multimodal model series by Google, supporting text and image inputs with text generation capabilities. This version is a 27B parameter instruction-tuned model using quantization-aware training, offering lower memory requirements while maintaining near-original quality.

Gemma 3 is a lightweight open model built by Google based on Gemini technology, supporting text and image inputs to generate text outputs. The 12B version is instruction-tuned and suitable for various generation and comprehension tasks.

Llama 4 Maverick is a 17-billion-parameter multimodal AI model developed by Meta, featuring a Mixture of Experts (MoE) architecture, supporting multilingual text and image understanding with 128 expert modules.

This model is a GGUF-format conversion of Qwen2.5-VL-7B-Instruct, supporting multimodal tasks and applicable to image and text interaction processing.

Gemma 3 is a lightweight cutting-edge open model family from Google, built on the same research technology as Gemini models. Supports text/image input and text output, offering both pretrained and instruction-tuned weight versions.

Gemma 3 is a lightweight multimodal open model from Google that supports text and image inputs with text outputs, featuring a 128K large context window and support for 140+ languages.

Sapnous-6B is an advanced vision-language model that enhances perception and understanding of the world through powerful multimodal capabilities.

OpenVLThinker-7B is a vision-language reasoning model specifically designed for multimodal tasks, with particular optimization for solving visual mathematical problems.

Gemma is Google's lightweight cutting-edge open-source model family, built using the same research technology as Gemini models. Gemma 3 is a multimodal model supporting text/image input and text output.

TimeZero is a reasoning-guided large-scale vision-language model (LVLM) specifically designed for temporal video grounding (TVG) tasks, achieving dynamic video-language relationship analysis through reinforcement learning methods.

TimeZero is a reasoning-guided large vision-language model (LVLM) specifically designed for temporal video grounding (TVG) tasks. It identifies temporal segments in videos corresponding to natural language queries through reinforcement learning methods.

Gemma 3 is a lightweight, advanced open model series launched by Google, built on the same research technology as Gemini, supporting multimodal input and text output.

Gemma 3 is a lightweight open model series launched by Google, built on the same technology as Gemini, supporting multimodal input and text output.

This is an MLX-converted version of the Google Gemma 3 27B IT model, supporting image-text-to-text tasks.