Fix license: correct license tag (was CC BY-NC-SA 4.0, actually Fish Audio Research License), add NOTICE
Browse files
NOTICE
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
This model is licensed under the Fish Audio Research License, Copyright © 39 AI, INC. All Rights Reserved.
|
| 2 |
+
|
| 3 |
+
This repository is a Derivative Work, under the Fish Audio Research License,
|
| 4 |
+
of fishaudio/s2-pro (https://huggingface.co/fishaudio/s2-pro).
|
| 5 |
+
|
| 6 |
+
Changes made:
|
| 7 |
+
- A LoRA adapter (r=32, alpha=16) on the attention and mlp modules of both the
|
| 8 |
+
slow and fast transformers.
|
| 9 |
+
- 26 atomic input-only tokens for Hebrew IPA symbols, each with its own
|
| 10 |
+
trainable embedding, added to the tokenizer (155,774 -> 155,800 tokens) and
|
| 11 |
+
trained alongside the LoRA weights. The output vocabulary is unchanged.
|
| 12 |
+
- A fix to an embedding-scale mismatch between the training and inference
|
| 13 |
+
paths in fish_speech/models/text2semantic/llama.py (upstream of this
|
| 14 |
+
adapter's own training), which applies to any S2-Pro fine-tune.
|
| 15 |
+
|
| 16 |
+
The full license text is at:
|
| 17 |
+
https://github.com/maxmelichov/fish-speech/blob/main/LICENSE
|