# Lightweight model

Gemma 3n-E2B-it is a lightweight open-source multimodal model launched by Google, built on the same technology as Gemini and optimized for low-resource devices.

Quantize the Menlo/Jan - nano model based on llama.cpp, providing model files of various quantization types to meet different hardware and performance requirements.

A dedicated text line recognition model for Devanagari in the PP-OCRv3_rec series developed by the PaddleOCR team, supporting Devanagari recognition with an average accuracy of 96.44%.

PP-LCNet_x1_0_table_cls is an efficient table classification model used to classify input table images and supports the classification of lined tables and borderless tables.

An ultra-lightweight Korean text line recognition model that supports the recognition of Korean and numeric characters, with an average accuracy of 60.21%.

An ultra-lightweight Latin text line recognition model developed by the PaddleOCR team, supporting the recognition of Latin and numeric characters.

An ultra-lightweight Arabic letter recognition model developed by the PaddleOCR team, supporting the recognition of Arabic letters and numeric characters.

Document image orientation classification module, used to distinguish the orientation of document images and correct them through post-processing to improve the accuracy of OCR processing.

A 500-million-parameter text generation model based on the Llama architecture, specifically designed for story creation.

BrtGPT-124M-Base is a basic model pre - trained on a large amount of English corpora and can be used for free. It solves the problems of cumbersome use of open - source models and high requirements for processing power.

This is a 4-bit quantized language model in MLX format, suitable for text generation tasks.

Devstral-Small-2505-bf16 is an MLX-format model converted from MistralAI's Devstral-Small-2505 model, supporting multiple language processing tasks.

FastVLM-0.5B-Stage3 is an efficient multimodal language model with visual understanding and language processing capabilities. It can process long videos and generate structured outputs.

FastVLM-0.5B-Stage2 is an efficient multimodal language model capable of understanding visual content and handling text tasks.

A lightweight language model developed by mistralai, optimized with 6-bit quantization via the MLX framework, specifically designed for Apple Silicon.

This is a sentence-transformers model fine-tuned from intfloat/multilingual-e5-small, trained with Korean query-passage pairs to enhance performance in Korean retrieval tasks.

MiniCOIL is a sparse contextualized word-by-word embedding model specifically designed for efficient semantic similarity computation

This model is distilled from all-MiniLM-L12-v2, achieving nearly 2x faster inference speed while maintaining high accuracy on both CPU and GPU.

Phi-4-mini-reasoning is a lightweight open model based on synthetic data, focusing on high-quality, dense reasoning data, and further fine-tuned to enhance mathematical reasoning capabilities.

This is a GGUF format model converted from kyutai/helium-1-2b, supporting multiple European languages.

Phi-4-mini-reasoning is a lightweight open-source model that focuses on high-quality, dense reasoning data and is further fine-tuned to obtain more advanced mathematical reasoning capabilities.

SegFormer image segmentation model based on NVIDIA MIT-B1 architecture, fine-tuned on specific datasets, excels in high-precision image segmentation tasks

A 1B-parameter language model specializing in English and Italian translation, offering multiple quantized versions in GGUF format.

A variant of the Gemma-2B model fused using the SLERP method, combining two different weighted versions of the Gemma-2B model

The ONNX version of TeapotLLM, supporting multilingual text generation models that run in the browser via Transformers.js.

Quantized version based on webbigdata/gemma-2-2b-jpn-it-translate, focusing on translation tasks between Japanese and English

This model is converted from ds4sd/SmolDocling-256M-preview to the MLX format, supporting image-text-to-text tasks.

This is an 8B parameter model based on the Llama-3 architecture, specializing in translation and processing of visual novels and Japanese otaku media content.

An image segmentation model fine-tuned based on nvidia/mit-b0, demonstrating outstanding performance on the evaluation set with a mean IoU of 0.9210 and mean accuracy of 0.9571.

GLiNER-Biomedical Edition is an efficient open NER model suite based on the GLiNER framework, specifically designed for the biomedical field, capable of identifying various entities in biomedical texts.

An automatic speech recognition model for informal Arabic, fine-tuned based on the Whisper Tiny architecture

A fast and accurate neural machine translation model for Chinese to English translation

DocFusion is a unified framework for document parsing tasks, aiming to solve the problems of system complexity and limited performance in existing document parsing methods and providing a more efficient and unified solution for document parsing.

A lightweight image classification model generated by HuggingPics for gender classification tasks.

An image classification model built based on the timm library, with zoobot-encoder-convnext_nano as the base model, capable of image classification.

Flex.1-alpha-Fp8 is the safetensors format version of the Flex.1-alpha model with float8_e4m3fn weights, suitable for text-to-image generation tasks.

A lightweight text ranking model suitable for information retrieval and document sorting tasks.

This is a minimalist sequence classification model built for TRL library unit testing

Lightweight image segmentation model based on MIT-B0 architecture, fine-tuned specifically for floor plan segmentation tasks

A BERT model fine-tuned based on google-bert/bert-base-uncased, with unspecified specific use