en.axm is NVIDIA's parakeet-tdt_ctc-110m speech recognition model
(https://huggingface.co/nvidia/parakeet-tdt_ctc-110m), licensed under the
Creative Commons Attribution 4.0 International license (CC-BY-4.0,
https://creativecommons.org/licenses/by/4.0/).

Changes made by AssistantX: only the CTC head is kept, BatchNorm is folded
into the convolutions, the query, key and value projections are merged into
one matrix, and the matrices are quantized (one scale per row) and reordered
for AssistantX's own inference engine: to 4 bits, rounded with the model's
own input statistics on a calibration set (no retraining), except the
subsampling's, which are quantized to 8 bits.
