Open Source PyTorch Projects
Discover 46 open source PyTorch repositories from GitHub, each with an analysis of what it does, key features, use cases and alternatives. PyTorch projects here are most often combined with Machine Learning, Python and Deep Learning. Last updated October 3, 2026.
46 repositories · updated October 3, 2026
optimum: Optimize Model Training and Inference on Target Hardware
Hugging Face Optimum adds tools for optimizing model training and inference across hardware backends. It suits teams using Transformers, Diffusers, TIMM, or Sentence Transformers who need hardware-specific deployment or training workflows.
litgpt: Train, Fine-Tune, and Deploy Large Language Models
LitGPT provides implementations and workflows for pretraining, fine-tuning, evaluating, and serving a range of large language models. It suits developers and researchers who want configurable training recipes and direct control over model code.
HunyuanVideo-Avatar: Create Audio-Driven Character Videos
HunyuanVideo-Avatar generates dynamic, emotion-controllable videos of one or more characters from avatar images and audio. It is aimed at creators and researchers who need expressive talking-avatar or dialogue video generation and have access to compatible NVIDIA GPU hardware.
CineScale: Generate High-Resolution Video Without Fine-Tuning
CineScale is an inference framework for generating high-resolution video with pretrained diffusion models, without fine-tuning. It targets researchers and practitioners who want to upscale generation beyond a model’s training resolution, including 4K workflows.
StreamDiffusion: Generate Images in Real Time with Diffusion
StreamDiffusion adapts diffusion pipelines for interactive image generation, with support for text-to-image and image-to-image workflows. It targets developers building responsive GPU-powered demos and applications.
picotron: Train Llama-Like Models with Distributed Parallelism
Picotron is a compact Python framework for learning and experimenting with distributed pre-training of Llama-like models. It demonstrates data, tensor, pipeline, and context parallelism, prioritizing readable code over peak performance.
DragGAN: Edit GAN Images by Dragging Points
DragGAN is a research tool for interactive point-based editing of images generated by StyleGAN models. It lets users move image features by dragging points while the model updates the image, making controlled edits easier than conventional prompt-based workflows.
SyncTalk: Generate Synchronized Talking-Head Videos
SyncTalk is a CVPR 2024 system for generating talking-head videos from a person’s footage and audio. It targets synchronized lip movement, facial expression, and head pose, with workflows for training on a subject or running inference with provided models.
vggt: Reconstruct 3D Scenes from Images
VGGT is a feed-forward vision model that estimates cameras, depth, point maps, and point tracks from one or more images. It suits researchers and developers who need fast 3D scene reconstruction without a traditional multi-stage pipeline.
LAM: Create Animatable 3D Gaussian Avatars from One Image
LAM reconstructs a 3D Gaussian head avatar from a single image and supports animation and rendering across devices. It is aimed at developers building digital humans, especially interactive avatars, and requires model assets and a compatible compute environment for local use.
multiresolution-time-series-transformer: Forecast Time Series at Multiple Scales
A PyTorch forecasting model that processes time series at several temporal resolutions, then fuses the representations to predict future values. It suits experimentation with multi-scale forecasting, with adaptations from the paper that users should account for.
PPS-Ctrl: Translate Colonoscopy Images for Depth Estimation
PPS-Ctrl explores controllable sim-to-real translation for colonoscopy images, using per-pixel shading maps to guide Stable Diffusion and ControlNet. The repository currently provides partial pseudocode rather than a complete, ready-to-run implementation.