shimmy
⚡ Pure-Rust WebGPU inference engine — OpenAI-API compatible, GGUF native, runs on any GPU. No Python. No llama.cpp. Single binary.
- 6k
- Rust
- Apache-2.0
- 9 d ago
⚡ Pure-Rust WebGPU inference engine — OpenAI-API compatible, GGUF native, runs on any GPU. No Python. No llama.cpp. Single binary.
A gopeed-extension for downloading models and datasets from huggingface, hf-mirror and modelscope. Huggingface download
Mount remote repositories, models and datasets managed by Git LFS instantly.
Multimodal-OCR is an experimental, high-performance visual reasoning and optical character recognition suite designed to accurately extract text, analyze visual content, and parse complex document structures. Built upon a diverse ecosystem of cutting-edge vision-language models.
Dataset abierto de modelos de HuggingFace: metadatos curados y normalizados, licencias clasificadas, parámetros reales, tumbas de modelos retirados y ficha técnica en castellano por modelo. Actualización incremental 3 veces al día.
Qwen3.8-27B-Object-Detection is a high-capacity vision-language grounding, object detection, and spatial path-mapping workspace powered by the Qwen/Qwen3.8-27B model. The pipeline utilizes native Flash Attention 2 kernels (kernels-community/flash-attn2@v3) to accelerate token decoding and high-resolution visual processing dedicated tasks.
Demonstration for the Qwen/Qwen-Image-Edit-2509 model, featuring lazy-loaded LoRA adapters for fast, specialized image edits like photo-to-anime conversion, angle changes, lighting restoration, skin editing, and upscaling.
QIE-Bbox-Studio (Qwen Image Edit Bounding Box Studio) is an advanced AI-powered image editing interface built on top of the Qwen2.5-VL and Qwen-Image-Edit models. This application allows users to manipulate images with extreme precision by defining bounding boxes and providing natural language prompts.
Multimodal-OCR3 is a highly capable, experimental optical character recognition and visual processing suite designed for precise text extraction, document parsing, and markdown generation. Leveraging a powerful selection of vision-language.
Qwen-Image-Edit-Object-Manipulator is an advanced, high-performance image manipulation workspace built on top of the Qwen/Qwen-Image-Edit-2511 base pipeline and enhanced by a custom optimized transformer.
Demonstration for NVIDIA's Nemotron-Parse-v1.1 model, designed for advanced document parsing and OCR. Upload images of documents (e.g., papers, forms) to extract structured content: text, tables (LaTeX), figures, and titles. Outputs annotated images with colored bounding boxes and processed markdown/LaTeX text for easy integration.
Qwen-3.5-HF-Demo is an experimental, advanced multimodal intelligence interface built on top of Alibaba Cloud's state-of-the-art Qwen/Qwen3.5-2B foundation model. Designed as a flexible multi-modal sandbox, this application unifies four complex spatial and temporal reasoning workflows: Image Understanding, Video Question Answering.
VLM-Parsing is a Gradio-based web application for parsing documents and images into structured HTML and Markdown formats using advanced Vision Language Models (VLMs).
🤗 A Python script for efficiently downloading and reconstructing large Hugging Face model files by splitting them into manageable chunks
Core-OCR is an advanced, experimental Optical Character Recognition (OCR) and document analysis suite designed for highly accurate text extraction, table reconstruction, and complex visual reasoning. Built on the robust Qwen2.5-VL and Qwen2-VL multimodal architectures.
Scribble-Board-Fast is an interactive, high-performance sketch-to-image synthesis workspace powered by the black-forest-labs/FLUX.2-klein-9B model. Built with Diffusers, it transforms freehand doodles, customized brush strokes, and modular visual assets into images.
This demonstrates the process of adapting a large scale pretrained model, MetaCLIP 2, for fine tuning a specific downstream task: image classification.
OpenCaption-4B-VL-SFT is an advanced multimodal image captioning terminal interface powered by prithivMLmods/OpenCaption-4B-VL-SFT-v1.0
Qwen3-VL-abliterated-MAX-Fast is an experimental, high-performance visual reasoning and optical character recognition (OCR) workspace. Powered by the unredacted prithivMLmods/Qwen3-VL-4B-Instruct-Unredacted-MAX architecture, this suite is designed to deliver uncensored, raw image-to-text processing.
A practical Large Language Model (LLM) fine-tuning repository focused on Low-Rank Adaptation (LoRA) and Parameter-Efficient Fine-Tuning (PEFT). Covers LoRA architecture, adapter configuration, dataset preparation, supervised fine-tuning, model evaluation, and efficient LLM customization using Python, PyTorch, and Hugging Face Transformers.
5 to 40 min. a day on a blackwell rtx 6000 for free or 9 $ month? Why not?
Every weight of google/gemma-3-1b-it updated — no adapters — to turn a general chat model into a dedicated paraphraser. iBLEU 17.99 against 1.65 zero-shot, with semantic similarity held at 0.911 and copy rate down from 41.7% to 0.7%. Trained with Unsloth on paraphrases.
HYZ-01-9B is a 9-billion-parameter open-source Large Language Model (LLM) developed by NeuroTurk AI. Optimized for high-performance text generation, reasoning, and multilingual tasks.
QLoRA fine-tune of Qwen3.5-0.8B on Bangla instruction data that did not work, published as a post-mortem. Four failure causes diagnosed, with benchmarks shown against the random-chance floor.
Recursive directory traversal with folder and file counts for Hugging Face models, spaces, and datasets.
A web application for viewing detailed statistics of any HuggingFace organization. Displays models, datasets, spaces, lifetime downloads, monthly downloads, and likes in a clean, tree-structured interface.
A web application for viewing detailed statistics of any HuggingFace user account. Displays models, datasets, spaces, lifetime downloads, monthly downloads, and likes in a clean, tree-structured interface.
Qwen-Image-Edit-2509-LoRAs-Fast-Fusion-Lazy-Load is an advanced multi-image generative studio and reference-based composition platform. Built upon the Qwen/Qwen-Image-Edit-2509 base architecture and an optimized transformer backend.
QIE-Object-Remover-Bbox-v3 is a highly advanced application for targeted object removal in images using bounding boxes. Built on the Gradio interface and powered by the latest Qwen Image Edit models.
A Gradio-based demonstration for the prithivMLmods/Gliese-CUA-Tool-Call-8B model, a Computer Use Agent (CUA) specialized in GUI understanding and tool-calling actions.
Demo: Herculis-CUA-GUI-Actioner-4B is a Computer Use Agent (CUA) multimodal model designed for GUI understanding, UI localization, and action execution across web, desktop, and mobile environments
A C++ project wrapper around a rich Web App for Qwen3.5 and Qwen3-VL models. Powered by pybind11 and an embedded native C++ HTTP server (httplib).
A C++ CLI tool for downloading, resharding, and re-uploading large Hugging Face models. It uses pybind11 to connect with Python libraries like transformers, huggingface_hub, and torch, enabling version conversion and configurable shard sizes.
Mergekit supports various architectures like Llama, Mistral, and more models, enabling merges on CPU or GPU with low memory needs through lazy tensor loading and out-of-core processing. It handles elaborate operations such as interpolated gradients, frankenmerging (layer-wise), LoRA extraction, and mixture of experts merging. [ by Arcee AI ]
Application for downloading, resharding, and re-uploading large Hugging Face models, with built-in optimizations for large Vision-Language (VL) models. It also maintains version control and enables the transformation of files from older Transformers versions to newer versions.