README: recommend the 0.666 VAE merge
Browse files
README.md
CHANGED
|
@@ -54,6 +54,12 @@ and `<name>.txt` = a style caption starting with your trigger phrase (e.g. `xyzq
|
|
| 54 |
~1,500 steps the model memorises the songs.
|
| 55 |
6. `LADDER_STYLE_TRACK=<song> LADDER_LYRICS=<lyrics.txt> bash ladder.sh my_lora nar_lora_joint_v4.pt` — renders one fixed prompt from every checkpoint (optional `FINALS=<file>` with lines `tag style_track lyrics seed`). Pick by ear (ours: step 800).
|
| 56 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 57 |
## Inference
|
| 58 |
```bash
|
| 59 |
# new song: style caption with your trigger + lyrics, score-free
|
|
|
|
| 54 |
~1,500 steps the model memorises the songs.
|
| 55 |
6. `LADDER_STYLE_TRACK=<song> LADDER_LYRICS=<lyrics.txt> bash ladder.sh my_lora nar_lora_joint_v4.pt` — renders one fixed prompt from every checkpoint (optional `FINALS=<file>` with lines `tag style_track lyrics seed`). Pick by ear (ours: step 800).
|
| 56 |
|
| 57 |
+
## Decoder
|
| 58 |
+
Consider decoding with [Mothersuperior/YuE2-Vae-merge-0.666](https://huggingface.co/Mothersuperior/YuE2-Vae-merge-0.666): a weight merge of
|
| 59 |
+
YuE2-Vae (0.666) and YuE2-Vae-legacy (0.334). The two releases share one encoder and only differ in the decoder; the merge sits between the
|
| 60 |
+
defaults perceptual cleanliness and the legacy decoders musicality. Drop-in: `YuE2Pipeline.from_pretrained("m-a-p/YuE2-3B", vae="Mothersuperior/YuE2-Vae-merge-0.666")`,
|
| 61 |
+
or point `ar_generate.py` at it.
|
| 62 |
+
|
| 63 |
## Inference
|
| 64 |
```bash
|
| 65 |
# new song: style caption with your trigger + lyrics, score-free
|