Time Series Forecasting with ci-proto: From Transformers to Mamba-3 on WebGPU
Porting Continuous Intelligence's Mamba-3 forecasting model to WebGPU with loop-free ONNX, Rust/WASM, and lazy multivariate loading.
Engineering notes, product updates, and practical thinking on local machine learning for structured data.
Porting Continuous Intelligence's Mamba-3 forecasting model to WebGPU with loop-free ONNX, Rust/WASM, and lazy multivariate loading.
How Carla runs Kumo Tabular Small in Chromium on a 16 GB laptop using GPU-resident caches and bounded-memory tiling.
From churn to grocery demand, see how structured data foundation models work in browser-based spreadsheets and what Carla can do today.
Run Timestral-123M directly in Google Sheets using WebGPU. Discover how Carla brings private, probabilistic time series forecasting to your browser.
Use Jev and System One models to turn free-form text into probabilistic semantic features for TabPFN, TabICL and other tabular foundation models.
Evaluating TabICLv2, TabPFN-3, TabFM, and EXAONE-Tabular on model weight size, latency, licensing, and WebGPU browser execution compatibility.
What explainable AI is, how Shapley Additive exPlanations work from first principles, and how to read SHAP waterfall and directional beeswarm plots.
How Carla uses Langevin diffusion and differential privacy to generate privacy-preserving synthetic data directly inside your spreadsheets.
How we navigated Google Cloud Trust & Safety OAuth verification for Carla.