Ecosystem & Stack: pytorch

Unsloth Studio provides a unified web interface for efficiently running and training open-source AI models locally across various operating systems and hardware.

A comprehensive framework from ModelScope for efficiently fine-tuning, evaluating, and deploying over 1000 large language models and multimodal large models using advanced techniques.

A high-performance serving framework designed to accelerate inference for large language models and multimodal AI models.

PEFT is a state-of-the-art library for Parameter-Efficient Fine-Tuning, drastically reducing the computational and storage costs of adapting large pretrained models.

A deep learning framework that simplifies PyTorch development by automating boilerplate engineering code, enabling scalable training from CPU to multi-node GPUs with minimal code changes.

A Python library for building and deploying high-performance AI model inference APIs and multi-model serving systems with ease.

A unified framework for building local, private, and secure enterprise RAG pipelines using small, specialized LLMs optimized for on-device and edge deployment.

A comprehensive, step-by-step guide and codebase for building a ChatGPT-like Large Language Model from scratch using PyTorch.

A tool for automatically removing censorship and safety alignment from transformer-based language models without expensive post-training.

A comprehensive, job-oriented guide for AI Agent development, covering core technologies, practical projects, and interview preparation.

A KVCache-centric disaggregated architecture for high-performance LLM serving, powering leading AI services.

An open-source, modern-design AI training tracking and visualization tool that integrates with 50+ mainstream frameworks, simplifying experiment management for AI teams.

An open collection of methodologies, tools, and step-by-step instructions for successfully training, fine-tuning, and inferencing large language and multi-modal models.

Deep Lake is an AI data runtime and database optimized for deep learning, offering multimodal data storage, querying, vector search, and streaming for LLM and deep learning applications.

A PyTorch library enabling accessible large language models by dramatically reducing memory consumption through k-bit quantization for both inference and training.

Ludwig is a low-code, declarative deep learning framework designed to simplify the building, training, and deployment of custom AI models, including LLMs and neural networks.

OpenVINO is an open-source toolkit designed to optimize and deploy deep learning models for efficient AI inference across diverse hardware platforms, from edge to cloud.

Colossal-AI makes training and deploying large AI models cheaper, faster, and more accessible through advanced distributed training techniques.

A lightweight library providing a vast hub of ready-to-use datasets and efficient tools for data manipulation in AI and machine learning workflows.

Towhee is a cutting-edge framework designed to simplify and accelerate neural data processing pipelines, particularly for unstructured multimodal data and LLM orchestration.

A Python library for representing, transmitting, storing, and retrieving multimodal data, designed for AI applications.

A PyTorch-based framework to adapt Meta AI's Segment Anything Model (SAM) for improved performance on challenging downstream computer vision tasks using adapters and prompts.

A toolkit for quantizing (compressing) Large Language Models (LLMs) with hardware acceleration across various GPUs and CPUs, integrating with popular inference frameworks.

A deep learning toolkit for training custom multimodal large language models focused on speech, language, audio, and music processing.

Firefly is an open-source toolkit for efficient large language model training, supporting pre-training, instruction fine-tuning, and DPO with methods like LoRA and QLoRA.

A comprehensive code repository accompanying a hands-on course for mastering Huggingface Transformers, covering fundamental concepts to advanced fine-tuning and deployment techniques.

A beginner-friendly UI for fine-tuning language models using LoRA on commodity NVIDIA GPUs, though the project is no longer actively maintained.

A cost-effective solution for fine-tuning ChatGLM-6B using LoRA, enabling personalized large language models.

An efficient toolkit for fine-tuning ChatGLM-6B models using PEFT methods, enabling customization and deployment of large language models.

A comprehensive Linux-based tutorial for deploying and fine-tuning open-source LLMs/MLLMs, tailored for Chinese beginners.

A unified library for parameter-efficient and modular transfer learning, extending HuggingFace Transformers with various adapter methods.

AirLLM optimizes large language model inference memory, enabling 70B LLMs on a single 4GB GPU without quantization, and 405B Llama3.1 on 8GB VRAM.

A comprehensive collection of PyTorch implementations for over 60 deep learning papers, accompanied by detailed side-by-side notes for enhanced understanding.

Provides Google Colab notebooks to easily run Stable Diffusion WebUI, including various models and extensions, though it is now outdated and superseded by TostUI.

A PyTorch library implementing LoRA (Low-Rank Adaptation) to efficiently fine-tune large language models by significantly reducing trainable parameters and storage requirements.

BELLE is an open-source project dedicated to fostering the development of Chinese conversational large language models, aiming to make LLMs accessible to everyone.

A tool for fast and efficient fine-tuning of diffusion models using Low-rank Adaptation (LoRA), producing small, shareable models.

YaYi is an open-source Chinese large language model series, built on LLaMA 2 & BLOOM, designed to provide secure, reliable, and domain-specific AI capabilities for enterprise customers through extensive multi-domain instruction tuning.

A human preference reward model for evaluating and improving text-to-image generation models.

A comprehensive collection of recipes and code for training various reward models crucial for Reinforcement Learning from Human Feedback (RLHF) in large language models.

An open-source framework providing code, models, and insights for stable Reinforcement Learning from Human Feedback (RLHF) training in Large Language Models, focusing on the PPO algorithm and reward modeling.

A modular PyTorch library for state-of-the-art diffusion models, enabling easy generation of images, audio, and more.

A free and open-source AI-powered tool for advanced image inpainting, outpainting, and object replacement using state-of-the-art models.

Integrates Stable Diffusion directly into Blender, enabling artists to generate textures, concept art, and 3D assets using text prompts without leaving their creative environment.

A PyTorch implementation of Dreamfusion, enabling text-to-3D and image-to-3D content generation using NeRF and Stable Diffusion.

A fast, minimal PyTorch port of DALL·E Mini, optimized for efficient text-to-image generation inference.

A PyTorch implementation of Google's Imagen, a state-of-the-art text-to-image neural network that surpasses DALL-E2 in synthesis quality.

A PyTorch implementation of OpenAI's DALL-E 2, enabling advanced text-to-image synthesis through a diffusion-based neural network architecture.

An open-source PyTorch implementation of OpenAI's DALL-E, a text-to-image transformer, including CLIP for generation ranking.

A scalable generative AI framework for building, customizing, and deploying models focused on Large Language Models, Multimodal, and Speech AI (ASR, TTS).

Silero Models offers a collection of pre-trained, end-to-end text-to-speech models designed for simplicity, speed, and natural-sounding speech generation.

An open-source Text-to-Speech system built on the VITS2 backbone, enhanced with multilingual BERT for improved speech synthesis.

Generate high-quality audiobooks and voiceovers from various text formats with synchronized captions.

A powerful open-source toolkit for real-time voice cloning and arbitrary speech generation from text.

Generate high-quality audiobooks in .m4b format from .epub e-books using advanced text-to-speech technology, with both command-line and graphical interfaces.

A powerful open-source web UI for few-shot voice conversion and text-to-speech, enabling high-quality voice cloning with minimal audio data.

A Dockerized FastAPI wrapper providing a high-performance, multi-platform (CPU/GPU) and multi-language API for the Kokoro-82M text-to-speech model, compatible with OpenAI's speech endpoint.

A declarative, transactional Python library for building multimodal AI applications with incremental data storage, transformation, indexing, and orchestration.

A unified, reproducible, and efficient multimodal evaluation toolkit for large language models across text, image, video, and audio tasks.

An open-source AI model family for high-fidelity, expressive speech and sound generation across diverse real-world applications.

A PyTorch implementation of BitNet, enabling highly efficient 1-bit transformers for large language models.

A modular PyTorch-based framework from Facebook AI Research for state-of-the-art vision and language multimodal AI research.

A Python library for extracting clean markdown, multimodal media, and structured data from complex documents using vision-language models.

A high-performance, comprehensive face recognition library built on PaddlePaddle and PyTorch.

UER-py is an open-source PyTorch-based framework for pre-training and fine-tuning NLP models, offering modularity, extensibility, and a comprehensive model zoo.

A comprehensive toolkit for fine-tuning ChatGLM-6B, ChatGLM2-6B, and ChatGLM3-6B models using various methods like Freeze, Lora, P-tuning, and full parameter fine-tuning.

A unified platform simplifying instruction-tuning, parameter-efficient methods, and large language model integration for researchers and developers.

Provides a local web interface and API for the ChatTTS model, enabling text-to-speech synthesis with support for mixed languages and numbers.

DiffSinger is an official PyTorch implementation of a singing voice synthesis (SVS) and text-to-speech (TTS) system, leveraging a shallow diffusion mechanism for high-quality audio generation.

A deep learning toolkit for Text-to-Speech, offering pretrained models, training tools, and dataset utilities.

An open-source, multi-voice, and prompt-controlled text-to-speech engine capable of generating speech with diverse emotions in English and Chinese.

StyleTTS 2 is a text-to-speech model that achieves human-level speech synthesis by leveraging style diffusion and adversarial training with large speech language models.

MetaVoice-1B is an open-source, 1.2B parameter foundational model for highly expressive, human-like text-to-speech synthesis and zero-shot voice cloning.

An open-source implementation of Microsoft's VALL-E X, enabling zero-shot multilingual text-to-speech synthesis and voice cloning with emotion control.

A deep learning library for advanced Text-to-Speech generation, offering high-quality speech synthesis with pretrained models and multi-language support.

A library implementing various parameter-efficient fine-tuning (PEFT) algorithms, including advanced LoRA variants, for Stable Diffusion models to enhance image generation.

Create dynamic videos by smoothly transitioning between text prompts using Stable Diffusion's latent space exploration.

An extensive node suite that integrates cutting-edge 3D generation algorithms and models into ComfyUI, enabling seamless processing of 3D inputs like meshes and UV textures.

Official SeedVR2 Video Upscaler for ComfyUI, enabling high-quality video and image upscaling, also runnable as a standalone CLI.