Open Models Almanac
Methodstraining objectiveauxiliary loss

single mention · training objective

Weighted averaging of audio codebook losses

Weighted averaging of audio codebook losses combines the per-codebook audio losses into a single loss via a weighted average.1 It assigns larger weights to low-index codebooks, emphasizing semantic and low-index acoustic codes, which empirically improves perceived audio quality.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. We combine the 8 audio codebook losses into a single via a weighted average — LFM2 Technical Report
  2. This emphasizes semantic and low-index acoustic codes in the loss, which we found empirically to improve perceived audio quality — LFM2 Technical Report

Also called Codebook weighting.

Filed under training objective › auxiliary loss