vivienhenz commited on
Commit
7ea5681
·
verified ·
1 Parent(s): 639f0ed

Upload folder using huggingface_hub

Browse files
Files changed (5) hide show
  1. README.md +54 -0
  2. config.json +32 -0
  3. model.pth +3 -0
  4. special_tokens.json +0 -0
  5. tokenizer.tiktoken +0 -0
README.md ADDED
@@ -0,0 +1,54 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - lb
4
+ tags:
5
+ - text-to-speech
6
+ - fish-speech
7
+ - luxembourgish
8
+ - tts
9
+ license: cc-by-nc-sa-4.0
10
+ datasets:
11
+ - denZLS/Luxembourgish-Male-TTS-for-LOD
12
+ base_model: fishaudio/openaudio-s1-mini
13
+ ---
14
+
15
+ # Fish Speech - Luxembourgish TTS
16
+
17
+ Fine-tuned [Fish Speech](https://github.com/fishaudio/fish-speech) Dual-AR model for Luxembourgish text-to-speech.
18
+
19
+ ## Model Details
20
+
21
+ - **Base Model:** [fishaudio/openaudio-s1-mini](https://huggingface.co/fishaudio/openaudio-s1-mini)
22
+ - **Architecture:** Dual-AR Transformer (860M parameters)
23
+ - **Language:** Luxembourgish (lb)
24
+ - **Training Data:** 32,000 samples from male Luxembourgish speaker
25
+ - **Training Steps:** 9,000 steps (~2.4 epochs)
26
+ - **Fine-tuned on:** NVIDIA RTX 5090
27
+
28
+ ## Usage
29
+
30
+ Requires [Fish Speech](https://github.com/fishaudio/fish-speech) installed.
31
+
32
+ ```bash
33
+ # WebUI
34
+ python tools/run_webui.py \
35
+ --llama-checkpoint-path vivienhenz/fish-speech-luxembourgish \
36
+ --decoder-checkpoint-path fishaudio/openaudio-s1-mini/codec.pth
37
+ ```
38
+
39
+ ## Training Details
40
+
41
+ - Dataset: 32,000 male voice samples (~28 hours)
42
+ - Optimizer: AdamW (lr=1e-4)
43
+ - Precision: bf16-mixed
44
+ - Training time: ~3 hours on RTX 5090
45
+
46
+ ## Example
47
+
48
+ Input: `d'nottär huet haut de mueren zwou venten.`
49
+
50
+ Output: Natural Luxembourgish male voice
51
+
52
+ ## License
53
+
54
+ CC-BY-NC-SA-4.0 (inherited from Fish Speech)
config.json ADDED
@@ -0,0 +1,32 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "attention_o_bias": false,
3
+ "attention_qk_norm": true,
4
+ "attention_qkv_bias": false,
5
+ "codebook_size": 4096,
6
+ "dim": 1024,
7
+ "dropout": 0.0,
8
+ "fast_attention_o_bias": false,
9
+ "fast_attention_qk_norm": false,
10
+ "fast_attention_qkv_bias": false,
11
+ "fast_dim": 1024,
12
+ "fast_head_dim": 64,
13
+ "fast_intermediate_size": 3072,
14
+ "fast_n_head": 16,
15
+ "fast_n_local_heads": 8,
16
+ "head_dim": 128,
17
+ "initializer_range": 0.03125,
18
+ "intermediate_size": 3072,
19
+ "max_seq_len": 8192,
20
+ "model_type": "dual_ar",
21
+ "n_fast_layer": 4,
22
+ "n_head": 16,
23
+ "n_layer": 28,
24
+ "n_local_heads": 8,
25
+ "norm_eps": 1e-06,
26
+ "num_codebooks": 10,
27
+ "rope_base": 1000000,
28
+ "scale_codebook_embeddings": true,
29
+ "tie_word_embeddings": false,
30
+ "use_gradient_checkpointing": true,
31
+ "vocab_size": 155776
32
+ }
model.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:006510c1f2d855aca83f6ad0e947fee94ea07be56d32a3f0783b4dfee4203c58
3
+ size 1720408503
special_tokens.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer.tiktoken ADDED
The diff for this file is too large to render. See raw diff