AI models directory for capability-aware discovery

UniDepth v2 (ViT-L/14) is a 0.4B-parameter PyTorch model for monocular metric depth estimation, leveraging a Vision Transformer backbone to predict scale-aware depth maps from single RGB images.

Qwen/Qwen3-8B is an 8-billion parameter large language model developed by Alibaba Cloud's Qwen team, available on Hugging Face under the Apache 2.0 license. It is designed for advanced text generation, conversational AI, and complex reasoning tasks, featuring support for tool use and long-context understanding.

T5-small is a compact, multilingual encoder-decoder transformer model developed by Google and hosted on Hugging Face. Licensed under Apache 2.0, it supports text-to-text generation tasks such as translation, summarization, and question answering across multiple languages including English, French, German, and Romanian. It is optimized for low-latency inference and efficient deployment on various hardware backends.

Qwen/Qwen3-0.6B is a lightweight, open-source large language model developed by Alibaba Cloud's Qwen team. Released under the permissive Apache 2.0 license, this 0.6-billion-parameter model is designed for high-efficiency deployment on consumer-grade hardware, including CPUs and low-end GPUs. It features a 32,768-token context window, multilingual capabilities, and native support for tool-use and function calling, making it an ideal candidate for edge computing, real-time chatbots, and resource-constrained environments.

LLaVA (Large Language and Vision Assistant) is an open-source, end-to-end trained multimodal model that combines a vision encoder with a large language model (Vicuna) to enable visual understanding and reasoning. Developed by researchers from Microsoft, the University of Wisconsin-Madison, and Columbia University, LLaVA allows users to interact with images through natural language, supporting tasks such as visual question answering, image captioning, and complex visual reasoning. The latest iteration, LLaVA 1.6, introduces significant improvements in input resolution, visual instruction tuning, and logical reasoning capabilities. It is widely accessible for local deployment via Ollama, offering privacy, offline functionality, and flexible hardware requirements across various model sizes.

BAAI/bge-small-en-v1.5 is a lightweight, high-performance sentence embedding model developed by the Beijing Academy of Artificial Intelligence (BAAI). Based on the BERT architecture with approximately 33 million parameters, it generates 384-dimensional vector representations optimized for English text. It is widely used for semantic search, retrieval-augmented generation (RAG), and clustering tasks, offering a balance between computational efficiency and accuracy under the permissive MIT License.

openai/clip-vit-base-patch32 is a foundational multimodal AI model developed by OpenAI and hosted on Hugging Face. It implements the Contrastive Language-Image Pre-training (CLIP) architecture, specifically utilizing a Vision Transformer (ViT-B/32) encoder paired with a text transformer. This model is designed to learn visual concepts directly from natural language supervision, enabling powerful zero-shot image classification and cross-modal retrieval capabilities without the need for task-specific training data.

timm/mobilenetv3_small_100.lamb_in1k is a highly efficient image classification model from the MobileNetV3 family, optimized for resource-constrained environments such as mobile devices and IoT hardware. Pretrained on ImageNet-1k using the LAMB optimizer, it offers a balance of accuracy and speed with approximately 2.5 million parameters.

BGE Reranker v2 M3 is a cross-encoder reranking model by BAAI, built on XLM-RoBERTa, supporting 100+ languages. It outputs relevance scores for query-document pairs, achieving state-of-the-art results on multilingual retrieval benchmarks. Licensed under Apache 2.0, it has over 16 million downloads and is deployable via Hugging Face, Sentence-Transformers, and Text Embeddings Inference.

nomic-embed-text is a high-performance, open-source text embedding model developed by Nomic AI. Designed for local deployment via Ollama, it supports a 2,048-token context window and generates 768-dimensional vectors. The model is licensed under Apache 2.0 and is optimized for semantic search, RAG pipelines, and document clustering, offering a privacy-preserving alternative to cloud-based embedding services.

Mistral is a 7.3 billion parameter open-weight language model developed by Mistral AI. Released under the permissive Apache 2.0 license, it is designed for high efficiency and strong performance on consumer-grade hardware. The model supports instruction following, coding, and multilingual tasks, with recent updates (v0.3) adding native function calling capabilities. It is widely accessible via the Ollama platform for local deployment.

sentence-transformers/all-mpnet-base-v2 is a high-performance, English-only sentence embedding model fine-tuned from Microsoft’s MPNet base architecture. It maps text into a 768-dimensional dense vector space, widely recognized for its state-of-the-art performance on semantic textual similarity (STS) and retrieval benchmarks. Licensed under Apache 2.0, it serves as a foundational component for semantic search, clustering, and information retrieval systems.

BAAI/bge-m3 is a state-of-the-art multilingual embedding model developed by the Beijing Academy of Artificial Intelligence (BAAI). Built on the XLM-RoBERTa architecture, it supports over 100 languages and offers three distinct retrieval modes: dense, sparse, and multi-vector (ColBERT-style). With a context window of up to 8192 tokens, it is optimized for long-document retrieval, cross-lingual search, and Retrieval-Augmented Generation (RAG) pipelines.

BERT Base Uncased is a foundational transformer model developed by Google, featuring 110 million parameters. It utilizes a bidirectional encoder architecture trained on BooksCorpus and English Wikipedia using Masked Language Modeling and Next Sentence Prediction. Licensed under Apache 2.0, it serves as a versatile backbone for tasks such as sentiment analysis, named entity recognition, and question answering, deployable across various frameworks including PyTorch, TensorFlow, and JAX.

sentence-transformers/all-MiniLM-L6-v2 is a highly efficient, open-source sentence embedding model based on the MiniLM architecture. Fine-tuned on over 1 billion sentence pairs, it maps text to a 384-dimensional vector space, making it ideal for semantic search, clustering, and similarity tasks. Licensed under Apache 2.0, it supports multiple deployment frameworks including Hugging Face Transformers, ONNX, and OpenVINO.

Qwen/Qwen3-4B is a 4-billion-parameter causal language model developed by Alibaba Cloud's Qwen team. Part of the Qwen3 series, it is optimized for text generation, code assistance, and multilingual tasks. Licensed under Apache 2.0, it supports efficient inference via libraries like Transformers, vLLM, and SGLang, and can run on consumer-grade hardware with 8GB+ VRAM.

Phi-3 is a family of lightweight, state-of-the-art open language models developed by Microsoft Research. Designed for efficiency and high performance, the family includes Phi-3 Mini (3.8B parameters) and Phi-3 Medium (14B parameters), with variants supporting context windows ranging from 4K to 128K tokens. These models are optimized for local deployment on edge devices, offering strong capabilities in reasoning, coding, and mathematics while maintaining a small footprint suitable for resource-constrained environments.

nomic-ai/nomic-embed-text-v1.5 is a high-performance, open-source text embedding model developed by Nomic AI. Built on the NomicBERT architecture, it is optimized for semantic search, clustering, and feature extraction tasks. The model stands out for its support of Matryoshka Representation Learning (MRL), allowing users to adjust embedding dimensions dynamically without retraining. It is released under the permissive Apache 2.0 license and is widely deployed via Hugging Face Transformers, sentence-transformers, ONNX Runtime, and Transformers.js.

Llama 3.3 is a 70-billion parameter multilingual large language model developed by Meta. It serves as a highly efficient alternative to the larger Llama 3.1 405B model, offering comparable performance while significantly reducing computational requirements. Optimized for instruction tuning and dialogue, it supports eight languages and features a 128K context window, making it suitable for complex reasoning, code generation, and multilingual applications.

Qwen2.5 is a comprehensive series of large language models developed by Alibaba Group's Tongyi Lab, available via Ollama for local deployment. It features parameter sizes ranging from 0.5B to 72B, supporting contexts up to 128K tokens and multilingual capabilities across 29+ languages. The series is distinguished by significant enhancements in coding, mathematics, instruction following, and structured output generation compared to its predecessor, Qwen2.