src/speech_decoding/models/v14_converged_v3/towers.py

Selected training excerpt · lines 370–374 · source SHA-256 5610fc7e8fc4

370    _, _, pred_d, pred_h = width_spec(enc_d_model)371    return V3Tower(372        PRED_LAYOUT, d_model=pred_d, n_heads=pred_h, n_parcels=n_parcels,373        parcel_embed=parcel_embed, space_rope=space_rope,374    )