WebGPU Local Inference

DevDiff v1.0.1 supports WebGPU-accelerated local model inference. This allows running lightweight models (e.g. Llama 3.2 3B, Gemma 2 2B) directly on your local GPU with zero installation and zero cloud costs.

Hardware Requirements

• GPU: A WebGPU-compatible GPU (e.g., Apple M-series chips, NVIDIA RTX GPUs, AMD Radeon GPUs).
• Driver support: Modern WebGPU backend support.

Fallback Chain

If WebGPU is unavailable, DevDiff automatically falls back to:

1. WebAssembly (WASM)
2. Native CPU
3. Local Ollama instance (with download instructions)