t0-alpha-q8_0-webgpu

Q8_0-quantized weights for t0-alpha, packaged for browser WebGPU forecasting.

theforecastingcompany/t0-alpha, by The Forecasting Company

Q8_0 quantization for WebGPU by Idle Intelligence.

Apache-2.0

Matches the official INT8 export’s size (about 109 MB) with less drift from the F32 reference (0.20% vs 0.79%) and about 2x its browser forecast speed (34.5 ms vs 64.0 ms, Apple M2). Costs +0.04% MASE and +0.02% CRPS against the F32 control on GIFT-Eval.

Type
model
Runs
browser, WASM, WebGPU
Size
~102M params, 108.9 MB (Q8_0)
Demo (GitHub Pages)
https://idle-intelligence.github.io/t0-web/web/
Source
idle-intelligence/t0-web
HuggingFace
idle-intelligence/t0-alpha-q8_0-webgpu
Perf
GIFT-Eval 97 configs, normalized to Seasonal Naive: MASE 0.7258, CRPS 0.4943
Status
maintained

← Home