Hugging Face's logo

syvai
/
tts-v0.1-finetuned

Text Generation
Transformers
Safetensors
llama
axolotl
Generated from Trainer
conversational
text-generation-inference
Model card Files Community
tts-v0.1-finetuned
File size: 4,156 Bytes 
a4b92b8
 
 
 
 
 
 
 
 
 
 
 
 
01eeb35
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a4b92b8
02c4ebe
 
a4b92b8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
 
---
library_name: transformers
base_model: syvai/tts-v1-pretrained
tags:
- axolotl
- generated_from_trainer
datasets:
- syvai/zac-coral-tts
model-index:
- name: tts-v1-finetuned
  results: []
---

# syv.ai TTS v0.1
TTS v0.1 er vores første open source tekst-til-tale model. Den er trænet på over 1000 timers dansk lyd.

## Model
Modellen er originalt en LLAMA 3.2 3B model, som er blevet trænet på 100.000 timers engelsk, og vi har efterfølgende trænet den til at tale dansk.

I det, at modellen er en LLM, så betyder det også, at der kan køres inferens på den ved hjælp af vLLM, ollama eller andre populære inferns-frameworks.

Vi anbefaler, at I kigger efter hvordan inferens er implementeret i [Orpheus](https://github.com/canopyai/Orpheus-TTS).

## Vi søger mere tale
Ligger du inde med lyd (gerne ikke oplæst), så hører vi gerne fra dig. Vi søger specifikt normal samtale lyd.

## Licens
Følger MIT for privatpersoner og organisationer der vil bruge modellen til forskning. Ved kommercielt brug skal der betales 1 kr. for en livstidslicens. Læs LICENSE.txt for den fulde licens.


## Træningskonfiguration

axolotl version: `0.8.0`
```yaml
base_model: syvai/tts-v1-pretrained
# Automatically upload checkpoint and final model to HF
hub_model_id: syvai/tts-v1-finetuned

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: true

datasets:
  - path: syvai/zac-coral-tts
    type: 
dataset_prepared_path: last_run_prepared
val_set_size: 0.01
eval_sample_packing: False
output_dir: ./outputs/finetuned

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

wandb_project: orph
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 4
num_epochs: 3
optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 2e-5

bf16: auto
tf32: false

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_steps: 3
evals_per_epoch: 5
saves_per_epoch: 5
weight_decay: 0.05

special_tokens:
  pad_token: <custom_token_7>

```

</details><br>

# tts-v1-finetuned

This model is a fine-tuned version of [syvai/tts-v1-pretrained](https://huggingface.co/syvai/tts-v1-pretrained) on the syvai/zac-coral-tts dataset.
It achieves the following results on the evaluation set:
- Loss: 4.2860

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 8
- total_train_batch_size: 32
- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 3
- num_epochs: 3.0

### Training results

| Training Loss | Epoch  | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| 4.9492        | 0.0246 | 1    | 4.8478          |
| 4.7181        | 0.1969 | 8    | 4.5872          |
| 4.5871        | 0.3938 | 16   | 4.4631          |
| 4.557         | 0.5908 | 24   | 4.3972          |
| 4.4965        | 0.7877 | 32   | 4.3521          |
| 4.4697        | 0.9846 | 40   | 4.3258          |
| 4.4525        | 1.1723 | 48   | 4.3083          |
| 4.4301        | 1.3692 | 56   | 4.2980          |
| 4.4459        | 1.5662 | 64   | 4.2915          |
| 4.4382        | 1.7631 | 72   | 4.2893          |
| 4.4315        | 1.96   | 80   | 4.2866          |
| 4.4178        | 2.1477 | 88   | 4.2861          |
| 4.4501        | 2.3446 | 96   | 4.2859          |
| 4.4121        | 2.5415 | 104  | 4.2856          |
| 4.4164        | 2.7385 | 112  | 4.2859          |
| 4.4264        | 2.9354 | 120  | 4.2860          |


### Framework versions

- Transformers 4.51.3
- Pytorch 2.6.0+cu124
- Datasets 3.5.0
- Tokenizers 0.21.1