|
|
--- |
|
|
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 |
|
|
|
