Expand description
TabFM inference on the Burn backend (Flex CPU).
Single-table inference: Fourier cell embedding, column SetTransformer
stacks, row interaction with RoPE, in-context transformer, decoder head,
with weights loaded straight from GGUF via zsfm_burn::GgufWeights.
Ensemble orchestration (ensemble::orchestrate) is host-level code that
calls predict_batch; porting it is a matter of swapping the model type
once this single-table path passes its gate.