75 repos across 3 sub-areas
Libraries and models for semantic and instance segmentation, mask generation, and related computer vision tasks. The cluster centers on PyTorch-based implementations with a strong emphasis on model hub integration via `pytorch_model_hub_mixin` and `safetensors` for efficient model distribution. Repositories here include foundational segmentation architectures, specialized models for mask prediction, and tools for working with segmented outputs, with some attention to autoregressive and discrete representation approaches for vision tasks.
PyTorch Model Hub & Serialization
28 repos
Libraries and tools for standardizing PyTorch model distribution, versioning, and storage using the model_hub_mixin pattern and safetensors format. This cluster focuses on infrastructure for sharing trained models across platforms—enabling reproducible model loading, safe serialization, and integration with Hugging Face Hub and similar model repositories. The central repos (OneEncoder variants, VGGT-1B, Pi3 series) demonstrate practical applications in multimodal vision tasks, but the cluster's cohesion comes primarily from shared adoption of PyTorch Hub conventions and safetensors as a common serialization standard.
PyTorch Model Hub & Safetensors Serialization
24 repos
Tools and infrastructure for publishing, managing, and serializing machine learning models using PyTorch's model hub ecosystem and the safetensors format. This cluster focuses on model standardization, safe weight storage, and hub integration patterns that make models portable and reproducible across platforms. The central repositories demonstrate dialect and language identification models (English, Mandarin, Cantonese) built on Whisper and MMS architectures, serving as reference implementations for how to package and distribute audio models through model hubs.
PyTorch Model Hub & Diffusion Policy
23 repos
Libraries and frameworks for publishing, managing, and distributing machine learning models through the Hugging Face Model Hub ecosystem, with a focus on PyTorch integration via the model_hub_mixin and safetensors standards. The cluster emphasizes reproducible model sharing and includes robotics applications through diffusion-based policy learning, as seen in projects like Switti and hma-base-disc that combine model distribution infrastructure with embodied AI use cases.