rust-candle-core

What problem does it solve?

Build native Rust ML models efficiently using the Candle framework, enabling GPU-accelerated pipelines for vision transformers, large language models, and audio processing.

Core Features & Use Cases

• GPU-accelerated Rust ML with Candle core crates including candle-core, candle-nn, and candle-transformers.
• Patterns for vision and language models such as PatchEmbed, multi-head attention, rotary position embeddings, and RMS normalization.
• Guidance on efficient loading with memory-mapped weights, CUDA feature flags, and practical examples for real-world ML workflows.

Quick Start

Initialize a Rust project configured with the Candle GPU workflow to begin building a simple Transformer-based model.