Card: w2 is [39x512], outputs by inputs
Browse files
README.md
CHANGED
|
@@ -28,8 +28,9 @@ sub_bass against bass_hit.
|
|
| 28 |
(LAION, Apache-2.0), converted to Core AI with its weights unchanged; it turns
|
| 29 |
the spectrogram into a 512-number embedding.
|
| 30 |
3. `head-*.f32` — a small trained head on the L2-normalised embedding:
|
| 31 |
-
512 → 512 (GELU) → 39, softmax. Little-endian float32
|
| 32 |
-
[512], `
|
|
|
|
| 33 |
|
| 34 |
Only the head was trained, on a licensed sample library; the audio is not
|
| 35 |
distributed and cannot be recovered from these files. Apache-2.0, following CLAP.
|
|
|
|
| 28 |
(LAION, Apache-2.0), converted to Core AI with its weights unchanged; it turns
|
| 29 |
the spectrogram into a 512-number embedding.
|
| 30 |
3. `head-*.f32` — a small trained head on the L2-normalised embedding:
|
| 31 |
+
512 → 512 (GELU) → 39, softmax. Little-endian float32, row-major in PyTorch's
|
| 32 |
+
Linear layout (outputs × inputs): `w1` [512×512], `b1` [512], `w2` [39×512],
|
| 33 |
+
`b2` [39].
|
| 34 |
|
| 35 |
Only the head was trained, on a licensed sample library; the audio is not
|
| 36 |
distributed and cannot be recovered from these files. Apache-2.0, following CLAP.
|