MossFormer2 SR 48K (MLX)
Audio super resolution, 16 kHz in to 48 kHz out. This model was converted to MLX
format from
alibabasglab/MossFormer2_SR_48K.
Files
| file | size |
|---|
model_fp32.safetensors | 418 MiB |
model_int8.safetensors | 293 MiB |
config.json | 392 B |
config.json is required: this model builds its architecture from the file rather
than hardcoding one, so a download without it cannot load.
Precision
| precision | size | speed | peak memory | quality vs fp32 |
|---|
| fp32 | 418 MiB | 2.1x | 4149 MiB | reference |
| int8 | 293 MiB | 2.1x | 3985 MiB | 62.5 dB |
Bold marks the best value in each column. Quality is int8 against this model's own
fp32 output — the cost of the precision change, not a score for the model.
Integer quantization reaches only the Linear modules, and this model's Generator
is almost entirely convolutions, so speed and memory are close to unchanged
between the two. int8 trades 30% of the download for 62.5 dB.
Measured on an Apple M1 Pro (16 GB), 30 s of audio, three timed runs per
configuration; speed is realtime factor. These are single-machine figures — expect
different absolute numbers on other hardware, and treat the ordering as indicative
rather than guaranteed.
Used by