Skill UI

High-quality Sentence Transformers framework that ships 5000+ pretrained models for multilingual, multimodal, and domain-specific embeddings; ideal for RAG, semantic search, clustering, and production vector stores without relying on external APIs.

Guides researchers through training and analyzing Sparse Autoencoders with SAELens to decompose neural activations into interpretable concepts, understand superposition, and steer or audit transformer models with sparse feature math.

Guides mechanistic interpretability research with TransformerLens by exposing HookPoints, activation caching, activation patching, and circuit-analysis workflows to inspect and manipulate GPT-style models for reverse-engineering attention behavior and causal tracing.

Load Hugging Face Transformers to run pipelines for text, vision, audio, and multimodal inference, customize model loading, tokenize inputs, and fine-tune with distributed training on domain-specific datasets.