lukasz-staniszewski commited on
Commit
f320676
·
verified ·
1 Parent(s): a4c9e53

Upload CAA steering vectors

Browse files
Files changed (3) hide show
  1. README.md +75 -0
  2. austeer.pkl +3 -0
  3. config.json +40 -0
README.md ADDED
@@ -0,0 +1,75 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: audio-interv
3
+ tags:
4
+ - ace-step
5
+ - activation-steering
6
+ - audio
7
+ - austeer
8
+ - diffusion
9
+ - interpretability
10
+ - music
11
+ - steering
12
+ - tempo
13
+ ---
14
+
15
+ # Steering Vectors — `tempo` (ACE-Step)
16
+
17
+ Steering vectors for the **tempo** concept on ACE-Step, computed via contrastive activation addition (CAA).
18
+
19
+ ## Quickstart
20
+
21
+ ```python
22
+ from src.steering import SteerableACEModel
23
+ from steering.caa import CAAController
24
+
25
+ model = SteerableACEModel(device="cuda")
26
+ ctrl = CAAController.from_pretrained("lukasz-staniszewski/ace-step-austeer-tempo-all", alpha=20)
27
+
28
+ with model.steer(ctrl):
29
+ audio = model.generate(prompt="instrumental music", seed=0)
30
+ ```
31
+
32
+ ## Generation config
33
+
34
+ ```json
35
+ {
36
+ "method": "austeer",
37
+ "concept": "tempo",
38
+ "lyrics": "[inst]",
39
+ "layers": "all",
40
+ "layers_collected": [
41
+ "tf0",
42
+ "tf1",
43
+ "tf2",
44
+ "tf3",
45
+ "tf4",
46
+ "tf5",
47
+ "tf6",
48
+ "tf7",
49
+ "tf8",
50
+ "tf9",
51
+ "tf10",
52
+ "tf11",
53
+ "tf12",
54
+ "tf13",
55
+ "tf14",
56
+ "tf15",
57
+ "tf16",
58
+ "tf17",
59
+ "tf18",
60
+ "tf19",
61
+ "tf20",
62
+ "tf21",
63
+ "tf22",
64
+ "tf23"
65
+ ],
66
+ "num_inference_steps": 30,
67
+ "audio_duration": 30.0,
68
+ "seed": 10,
69
+ "guidance_scale": 5.0,
70
+ "guidance_scale_text": 0.0,
71
+ "guidance_scale_lyric": 0.0,
72
+ "guidance_interval": 1.0,
73
+ "guidance_interval_decay": 0.0
74
+ }
75
+ ```
austeer.pkl ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:feb0212c469bf15eb853bad5c1234775d9902818151f64aa26dfdc281106f5c9
3
+ size 29551625
config.json ADDED
@@ -0,0 +1,40 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "method": "austeer",
3
+ "concept": "tempo",
4
+ "lyrics": "[inst]",
5
+ "layers": "all",
6
+ "layers_collected": [
7
+ "tf0",
8
+ "tf1",
9
+ "tf2",
10
+ "tf3",
11
+ "tf4",
12
+ "tf5",
13
+ "tf6",
14
+ "tf7",
15
+ "tf8",
16
+ "tf9",
17
+ "tf10",
18
+ "tf11",
19
+ "tf12",
20
+ "tf13",
21
+ "tf14",
22
+ "tf15",
23
+ "tf16",
24
+ "tf17",
25
+ "tf18",
26
+ "tf19",
27
+ "tf20",
28
+ "tf21",
29
+ "tf22",
30
+ "tf23"
31
+ ],
32
+ "num_inference_steps": 30,
33
+ "audio_duration": 30.0,
34
+ "seed": 10,
35
+ "guidance_scale": 5.0,
36
+ "guidance_scale_text": 0.0,
37
+ "guidance_scale_lyric": 0.0,
38
+ "guidance_interval": 1.0,
39
+ "guidance_interval_decay": 0.0
40
+ }