🧠 Meta-Planner (OpenForecaster-8B)

Meta-planning + calibrated forecasting with selectable quantization:

  • GGUF: IQ2_XXS, IQ3_XXS, IQ4_XS
  • BF16 (GPU/ZeroGPU required)

Enable Auto-save confident predictions to build a persistent bets ledger.

Ready.
Compute mode
Quantization
0 1.5
32 1024
0 1

Saved Bets

No bets saved yet.

Browser WebGPU mode (experimental)

This mode keeps compute on the visitor's machine, but OpenForecaster-8B currently does not have a public ONNX/WebGPU package. So this Space cannot run full OpenForecaster-8B in browser WebGPU yet.

Try the in-Space browser client here:

What works today in this Space:

  • Server mode with GGUF quants: IQ2_XXS / IQ3_XXS / IQ4_XS
  • Server mode with BF16 when GPU/ZeroGPU is available

How to enable true browser WebGPU for Meta-Planner later:

  1. Export OpenForecaster-8B to an ONNX/WebGPU-compatible model repo.
  2. Point this mode to that ONNX model.