Upload AudioLDM2 CAA steering vectors (piano, all attn2, 100 steps, gs=4.5)
Browse files- README.md +117 -0
- config.json +80 -0
- neg_vectors.pkl +3 -0
- pos_vectors.pkl +3 -0
- sv.pkl +3 -0
README.md
ADDED
|
@@ -0,0 +1,117 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: audio-interv
|
| 3 |
+
tags:
|
| 4 |
+
- ace-step
|
| 5 |
+
- activation-steering
|
| 6 |
+
- audio
|
| 7 |
+
- audioldm
|
| 8 |
+
- audioldm2
|
| 9 |
+
- caa
|
| 10 |
+
- diffusion
|
| 11 |
+
- interpretability
|
| 12 |
+
- music
|
| 13 |
+
- piano
|
| 14 |
+
- steering
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# Steering Vectors — `piano` (ACE-Step)
|
| 18 |
+
|
| 19 |
+
Steering vectors for the **piano** concept on ACE-Step, computed via contrastive activation addition (CAA).
|
| 20 |
+
|
| 21 |
+
## Quickstart
|
| 22 |
+
|
| 23 |
+
```python
|
| 24 |
+
from src.steering import SteerableACEModel
|
| 25 |
+
from src.steering.methods.caa import CAASteeringController
|
| 26 |
+
|
| 27 |
+
model = SteerableACEModel(device="cuda")
|
| 28 |
+
ctrl = CAASteeringController.from_pretrained("lukasz-staniszewski/audioldm2-caa-piano", alpha=20)
|
| 29 |
+
|
| 30 |
+
with model.steer(ctrl):
|
| 31 |
+
audio = model.generate(prompt="instrumental music", seed=0)
|
| 32 |
+
```
|
| 33 |
+
|
| 34 |
+
## Generation config
|
| 35 |
+
|
| 36 |
+
```json
|
| 37 |
+
{
|
| 38 |
+
"method": "standard_caa_audioldm",
|
| 39 |
+
"model": "cvssp/audioldm2-large",
|
| 40 |
+
"concept": "piano",
|
| 41 |
+
"num_inference_steps": 100,
|
| 42 |
+
"audio_length_in_s": 10.0,
|
| 43 |
+
"guidance_scale": 4.5,
|
| 44 |
+
"seed": 10,
|
| 45 |
+
"device": "cuda",
|
| 46 |
+
"dtype": "float16",
|
| 47 |
+
"save_all_cfg_passes": true,
|
| 48 |
+
"layers_preset": "all",
|
| 49 |
+
"layers_to_steer": [
|
| 50 |
+
".unet.down_blocks.1.attentions.1.transformer_blocks.0.attn2",
|
| 51 |
+
".unet.down_blocks.1.attentions.1.transformer_blocks.1.attn2",
|
| 52 |
+
".unet.down_blocks.1.attentions.2.transformer_blocks.0.attn2",
|
| 53 |
+
".unet.down_blocks.1.attentions.2.transformer_blocks.1.attn2",
|
| 54 |
+
".unet.down_blocks.1.attentions.5.transformer_blocks.0.attn2",
|
| 55 |
+
".unet.down_blocks.1.attentions.5.transformer_blocks.1.attn2",
|
| 56 |
+
".unet.down_blocks.1.attentions.6.transformer_blocks.0.attn2",
|
| 57 |
+
".unet.down_blocks.1.attentions.6.transformer_blocks.1.attn2",
|
| 58 |
+
".unet.down_blocks.2.attentions.1.transformer_blocks.0.attn2",
|
| 59 |
+
".unet.down_blocks.2.attentions.1.transformer_blocks.1.attn2",
|
| 60 |
+
".unet.down_blocks.2.attentions.2.transformer_blocks.0.attn2",
|
| 61 |
+
".unet.down_blocks.2.attentions.2.transformer_blocks.1.attn2",
|
| 62 |
+
".unet.down_blocks.2.attentions.5.transformer_blocks.0.attn2",
|
| 63 |
+
".unet.down_blocks.2.attentions.5.transformer_blocks.1.attn2",
|
| 64 |
+
".unet.down_blocks.2.attentions.6.transformer_blocks.0.attn2",
|
| 65 |
+
".unet.down_blocks.2.attentions.6.transformer_blocks.1.attn2",
|
| 66 |
+
".unet.down_blocks.3.attentions.1.transformer_blocks.0.attn2",
|
| 67 |
+
".unet.down_blocks.3.attentions.1.transformer_blocks.1.attn2",
|
| 68 |
+
".unet.down_blocks.3.attentions.2.transformer_blocks.0.attn2",
|
| 69 |
+
".unet.down_blocks.3.attentions.2.transformer_blocks.1.attn2",
|
| 70 |
+
".unet.down_blocks.3.attentions.5.transformer_blocks.0.attn2",
|
| 71 |
+
".unet.down_blocks.3.attentions.5.transformer_blocks.1.attn2",
|
| 72 |
+
".unet.down_blocks.3.attentions.6.transformer_blocks.0.attn2",
|
| 73 |
+
".unet.down_blocks.3.attentions.6.transformer_blocks.1.attn2",
|
| 74 |
+
".unet.mid_block.attentions.1.transformer_blocks.0.attn2",
|
| 75 |
+
".unet.mid_block.attentions.1.transformer_blocks.1.attn2",
|
| 76 |
+
".unet.mid_block.attentions.2.transformer_blocks.0.attn2",
|
| 77 |
+
".unet.mid_block.attentions.2.transformer_blocks.1.attn2",
|
| 78 |
+
".unet.up_blocks.0.attentions.1.transformer_blocks.0.attn2",
|
| 79 |
+
".unet.up_blocks.0.attentions.1.transformer_blocks.1.attn2",
|
| 80 |
+
".unet.up_blocks.0.attentions.2.transformer_blocks.0.attn2",
|
| 81 |
+
".unet.up_blocks.0.attentions.2.transformer_blocks.1.attn2",
|
| 82 |
+
".unet.up_blocks.0.attentions.5.transformer_blocks.0.attn2",
|
| 83 |
+
".unet.up_blocks.0.attentions.5.transformer_blocks.1.attn2",
|
| 84 |
+
".unet.up_blocks.0.attentions.6.transformer_blocks.0.attn2",
|
| 85 |
+
".unet.up_blocks.0.attentions.6.transformer_blocks.1.attn2",
|
| 86 |
+
".unet.up_blocks.0.attentions.9.transformer_blocks.0.attn2",
|
| 87 |
+
".unet.up_blocks.0.attentions.9.transformer_blocks.1.attn2",
|
| 88 |
+
".unet.up_blocks.0.attentions.10.transformer_blocks.0.attn2",
|
| 89 |
+
".unet.up_blocks.0.attentions.10.transformer_blocks.1.attn2",
|
| 90 |
+
".unet.up_blocks.1.attentions.1.transformer_blocks.0.attn2",
|
| 91 |
+
".unet.up_blocks.1.attentions.1.transformer_blocks.1.attn2",
|
| 92 |
+
".unet.up_blocks.1.attentions.2.transformer_blocks.0.attn2",
|
| 93 |
+
".unet.up_blocks.1.attentions.2.transformer_blocks.1.attn2",
|
| 94 |
+
".unet.up_blocks.1.attentions.5.transformer_blocks.0.attn2",
|
| 95 |
+
".unet.up_blocks.1.attentions.5.transformer_blocks.1.attn2",
|
| 96 |
+
".unet.up_blocks.1.attentions.6.transformer_blocks.0.attn2",
|
| 97 |
+
".unet.up_blocks.1.attentions.6.transformer_blocks.1.attn2",
|
| 98 |
+
".unet.up_blocks.1.attentions.9.transformer_blocks.0.attn2",
|
| 99 |
+
".unet.up_blocks.1.attentions.9.transformer_blocks.1.attn2",
|
| 100 |
+
".unet.up_blocks.1.attentions.10.transformer_blocks.0.attn2",
|
| 101 |
+
".unet.up_blocks.1.attentions.10.transformer_blocks.1.attn2",
|
| 102 |
+
".unet.up_blocks.2.attentions.1.transformer_blocks.0.attn2",
|
| 103 |
+
".unet.up_blocks.2.attentions.1.transformer_blocks.1.attn2",
|
| 104 |
+
".unet.up_blocks.2.attentions.2.transformer_blocks.0.attn2",
|
| 105 |
+
".unet.up_blocks.2.attentions.2.transformer_blocks.1.attn2",
|
| 106 |
+
".unet.up_blocks.2.attentions.5.transformer_blocks.0.attn2",
|
| 107 |
+
".unet.up_blocks.2.attentions.5.transformer_blocks.1.attn2",
|
| 108 |
+
".unet.up_blocks.2.attentions.6.transformer_blocks.0.attn2",
|
| 109 |
+
".unet.up_blocks.2.attentions.6.transformer_blocks.1.attn2",
|
| 110 |
+
".unet.up_blocks.2.attentions.9.transformer_blocks.0.attn2",
|
| 111 |
+
".unet.up_blocks.2.attentions.9.transformer_blocks.1.attn2",
|
| 112 |
+
".unet.up_blocks.2.attentions.10.transformer_blocks.0.attn2",
|
| 113 |
+
".unet.up_blocks.2.attentions.10.transformer_blocks.1.attn2"
|
| 114 |
+
],
|
| 115 |
+
"normalize_sv": true
|
| 116 |
+
}
|
| 117 |
+
```
|
config.json
ADDED
|
@@ -0,0 +1,80 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"method": "standard_caa_audioldm",
|
| 3 |
+
"model": "cvssp/audioldm2-large",
|
| 4 |
+
"concept": "piano",
|
| 5 |
+
"num_inference_steps": 100,
|
| 6 |
+
"audio_length_in_s": 10.0,
|
| 7 |
+
"guidance_scale": 4.5,
|
| 8 |
+
"seed": 10,
|
| 9 |
+
"device": "cuda",
|
| 10 |
+
"dtype": "float16",
|
| 11 |
+
"save_all_cfg_passes": true,
|
| 12 |
+
"layers_preset": "all",
|
| 13 |
+
"layers_to_steer": [
|
| 14 |
+
".unet.down_blocks.1.attentions.1.transformer_blocks.0.attn2",
|
| 15 |
+
".unet.down_blocks.1.attentions.1.transformer_blocks.1.attn2",
|
| 16 |
+
".unet.down_blocks.1.attentions.2.transformer_blocks.0.attn2",
|
| 17 |
+
".unet.down_blocks.1.attentions.2.transformer_blocks.1.attn2",
|
| 18 |
+
".unet.down_blocks.1.attentions.5.transformer_blocks.0.attn2",
|
| 19 |
+
".unet.down_blocks.1.attentions.5.transformer_blocks.1.attn2",
|
| 20 |
+
".unet.down_blocks.1.attentions.6.transformer_blocks.0.attn2",
|
| 21 |
+
".unet.down_blocks.1.attentions.6.transformer_blocks.1.attn2",
|
| 22 |
+
".unet.down_blocks.2.attentions.1.transformer_blocks.0.attn2",
|
| 23 |
+
".unet.down_blocks.2.attentions.1.transformer_blocks.1.attn2",
|
| 24 |
+
".unet.down_blocks.2.attentions.2.transformer_blocks.0.attn2",
|
| 25 |
+
".unet.down_blocks.2.attentions.2.transformer_blocks.1.attn2",
|
| 26 |
+
".unet.down_blocks.2.attentions.5.transformer_blocks.0.attn2",
|
| 27 |
+
".unet.down_blocks.2.attentions.5.transformer_blocks.1.attn2",
|
| 28 |
+
".unet.down_blocks.2.attentions.6.transformer_blocks.0.attn2",
|
| 29 |
+
".unet.down_blocks.2.attentions.6.transformer_blocks.1.attn2",
|
| 30 |
+
".unet.down_blocks.3.attentions.1.transformer_blocks.0.attn2",
|
| 31 |
+
".unet.down_blocks.3.attentions.1.transformer_blocks.1.attn2",
|
| 32 |
+
".unet.down_blocks.3.attentions.2.transformer_blocks.0.attn2",
|
| 33 |
+
".unet.down_blocks.3.attentions.2.transformer_blocks.1.attn2",
|
| 34 |
+
".unet.down_blocks.3.attentions.5.transformer_blocks.0.attn2",
|
| 35 |
+
".unet.down_blocks.3.attentions.5.transformer_blocks.1.attn2",
|
| 36 |
+
".unet.down_blocks.3.attentions.6.transformer_blocks.0.attn2",
|
| 37 |
+
".unet.down_blocks.3.attentions.6.transformer_blocks.1.attn2",
|
| 38 |
+
".unet.mid_block.attentions.1.transformer_blocks.0.attn2",
|
| 39 |
+
".unet.mid_block.attentions.1.transformer_blocks.1.attn2",
|
| 40 |
+
".unet.mid_block.attentions.2.transformer_blocks.0.attn2",
|
| 41 |
+
".unet.mid_block.attentions.2.transformer_blocks.1.attn2",
|
| 42 |
+
".unet.up_blocks.0.attentions.1.transformer_blocks.0.attn2",
|
| 43 |
+
".unet.up_blocks.0.attentions.1.transformer_blocks.1.attn2",
|
| 44 |
+
".unet.up_blocks.0.attentions.2.transformer_blocks.0.attn2",
|
| 45 |
+
".unet.up_blocks.0.attentions.2.transformer_blocks.1.attn2",
|
| 46 |
+
".unet.up_blocks.0.attentions.5.transformer_blocks.0.attn2",
|
| 47 |
+
".unet.up_blocks.0.attentions.5.transformer_blocks.1.attn2",
|
| 48 |
+
".unet.up_blocks.0.attentions.6.transformer_blocks.0.attn2",
|
| 49 |
+
".unet.up_blocks.0.attentions.6.transformer_blocks.1.attn2",
|
| 50 |
+
".unet.up_blocks.0.attentions.9.transformer_blocks.0.attn2",
|
| 51 |
+
".unet.up_blocks.0.attentions.9.transformer_blocks.1.attn2",
|
| 52 |
+
".unet.up_blocks.0.attentions.10.transformer_blocks.0.attn2",
|
| 53 |
+
".unet.up_blocks.0.attentions.10.transformer_blocks.1.attn2",
|
| 54 |
+
".unet.up_blocks.1.attentions.1.transformer_blocks.0.attn2",
|
| 55 |
+
".unet.up_blocks.1.attentions.1.transformer_blocks.1.attn2",
|
| 56 |
+
".unet.up_blocks.1.attentions.2.transformer_blocks.0.attn2",
|
| 57 |
+
".unet.up_blocks.1.attentions.2.transformer_blocks.1.attn2",
|
| 58 |
+
".unet.up_blocks.1.attentions.5.transformer_blocks.0.attn2",
|
| 59 |
+
".unet.up_blocks.1.attentions.5.transformer_blocks.1.attn2",
|
| 60 |
+
".unet.up_blocks.1.attentions.6.transformer_blocks.0.attn2",
|
| 61 |
+
".unet.up_blocks.1.attentions.6.transformer_blocks.1.attn2",
|
| 62 |
+
".unet.up_blocks.1.attentions.9.transformer_blocks.0.attn2",
|
| 63 |
+
".unet.up_blocks.1.attentions.9.transformer_blocks.1.attn2",
|
| 64 |
+
".unet.up_blocks.1.attentions.10.transformer_blocks.0.attn2",
|
| 65 |
+
".unet.up_blocks.1.attentions.10.transformer_blocks.1.attn2",
|
| 66 |
+
".unet.up_blocks.2.attentions.1.transformer_blocks.0.attn2",
|
| 67 |
+
".unet.up_blocks.2.attentions.1.transformer_blocks.1.attn2",
|
| 68 |
+
".unet.up_blocks.2.attentions.2.transformer_blocks.0.attn2",
|
| 69 |
+
".unet.up_blocks.2.attentions.2.transformer_blocks.1.attn2",
|
| 70 |
+
".unet.up_blocks.2.attentions.5.transformer_blocks.0.attn2",
|
| 71 |
+
".unet.up_blocks.2.attentions.5.transformer_blocks.1.attn2",
|
| 72 |
+
".unet.up_blocks.2.attentions.6.transformer_blocks.0.attn2",
|
| 73 |
+
".unet.up_blocks.2.attentions.6.transformer_blocks.1.attn2",
|
| 74 |
+
".unet.up_blocks.2.attentions.9.transformer_blocks.0.attn2",
|
| 75 |
+
".unet.up_blocks.2.attentions.9.transformer_blocks.1.attn2",
|
| 76 |
+
".unet.up_blocks.2.attentions.10.transformer_blocks.0.attn2",
|
| 77 |
+
".unet.up_blocks.2.attentions.10.transformer_blocks.1.attn2"
|
| 78 |
+
],
|
| 79 |
+
"normalize_sv": true
|
| 80 |
+
}
|
neg_vectors.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3089bdd8bb7e6cbae5b5385fce4da853646449624ff4a5ad36b5145aa4241909
|
| 3 |
+
size 1152639741
|
pos_vectors.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:557a12324f9def81b93382bfd7380f0ba20f686f19e430bb04ef30a54ff7a498
|
| 3 |
+
size 1152639741
|
sv.pkl
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8a98937b4d442e447731d68c3606da80b625316df5ce1dc1781906de466ef414
|
| 3 |
+
size 23058335
|