MiniMax Music 3 Text Encoder — W4A8
Pruned and quantized Text Encoder for MiniMax Music 3.
Audio Comparison
The following samples compare DIT INT8 and DIT FP16 using three different Text Encoder precisions: W4A8, INT8, and BF16.
| TE | DIT INT8 | DIT FP16 |
|---|---|---|
| W4A8 | 450.58 s |
609 s |
| INT8 | 896.84 s |
1013.56 s |
| BF16 | 2098.38 s |
2198.62 s |
Model
minimax_music3_text_encoder_pruned_w4a8.safetensors
Repository Contents
The repository contains the quantized Text Encoder and the script used to create it.
License
MiniMax Music 3 Community License
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support