What is AMEVA-Forge?
AMEVA-Forge is a browser-first deep learning acceleration framework designed to democratize machine learning education, research, and client-side privacy-preserving computation.
Core Philosophy
- Zero Installation Barrier: Traditional deep learning setups require gigabytes of CUDA drivers and PyTorch binaries. Forge runs instantly inside any modern browser with a single URL visit.
- Absolute Privacy: By executing forward and backward training passes on the client device's GPU, user data never traverses external network boundaries.
- Mathematical Parity: Every kernel and backward step is verified against CPU analytical ground truth to ensure absolute numerical correctness.
System Architecture
- Frontend (Python / Pyodide): Intercepts user model code (e.g.
nn.Linear,F.scaled_dot_product_attention) and constructs computation graphs. - Bridge Layer (JS / WASM): Serializes graph instructions with shape validation and memory quota pre-checking.
- Backend Engine (TypeScript / WebGPU): Compiles instructions into optimized WGSL shader dispatches using 8D uniform strides and pooled buffers.