src/speech_decoding/models/v14_converged_v3/objective.py

Selected training excerpt · lines 514–518 · historical width comment corrected to the released width · source SHA-256 5718458b537d

514            pred_in = pred_in + self.pred_band_emb[grid.band][None]  # (B, total, d_pred)515        # forward_flat adds the parcel embed to every token itself (towers._run_flat), so with the516        # band embed above the predictor now sees parcel + band on ALL tokens — the same identity517        # pair the encoder gets before its first L1 block.518        h = self.predictor.forward_flat(pred_in, grid, parcel_packed)  # (B, total, 192)