Show HN: Prompt-to-Excalidraw demo with Gemma 4 E2B in the browser (3.1GB)
5 hours ago
- #TurboQuant
- #WebGPU
- #Excalidraw
- Gemma 4 E2B enables describing diagrams and generating them as Excalidraw code entirely in the browser, requiring Desktop Chrome 134+ and producing compact code.
- The TurboQuant algorithm compresses the KV cache by approximately 2.4x to allow longer conversations in GPU memory, using WGSL compute shaders for GPU execution and a WASM+SIMD implementation for CPU-side vector search.
- Technical requirements include WebGPU subgroups (excluding Safari/iOS) and around 3 GB of RAM, with mobile browsers not meeting the memory requirements for this demo.