mirror of
https://github.com/Nighthawk42/llm-tts-factory.git
synced 2026-08-30 07:22:27 +00:00
116 lines
3.6 KiB
YAML
116 lines
3.6 KiB
YAML
# ==============================================================================
|
|
# LLM-TTS-Factory Configuration
|
|
# ==============================================================================
|
|
|
|
# ------------------------------------------------------------------------------
|
|
# Global Settings
|
|
# ------------------------------------------------------------------------------
|
|
global:
|
|
seed: 1337
|
|
device: "cuda:0"
|
|
num_workers: 4
|
|
use_wandb: true
|
|
wandb_project: "soprano-tts"
|
|
tokenizer_name: "ekwek/Soprano-80M"
|
|
|
|
# ------------------------------------------------------------------------------
|
|
# File Paths
|
|
# Use forward slashes (/) for both Windows and Linux, or standard OS paths.
|
|
# Relative paths are evaluated from the directory where the script is run.
|
|
#
|
|
# Examples:
|
|
# Linux: "/home/ubuntu/data/lj_speech/LJSpeech-1.1"
|
|
# Windows: "C:/Users/Name/Documents/datasets/LJSpeech-1.1"
|
|
# Relative: "./data/LJSpeech-1.1"
|
|
# ------------------------------------------------------------------------------
|
|
paths:
|
|
# Path to LJSpeech-1.1 directory
|
|
dataset_root: "./data/LJSpeech-1.1"
|
|
|
|
# Base directory to save all checkpoints, logs, and generated datasets
|
|
save_dir: "./checkpoints"
|
|
|
|
# Pretrained model paths (set to null if training from scratch or not applicable)
|
|
pretrained_codec_path: null
|
|
pretrained_llm_path: null
|
|
pretrained_decoder_path: null
|
|
pretrained_discriminator_path: null
|
|
|
|
# ------------------------------------------------------------------------------
|
|
# Data Generation Configuration (generate_dataset*.py)
|
|
# ------------------------------------------------------------------------------
|
|
data_generation:
|
|
val_prop: 0.1
|
|
val_max: 512
|
|
|
|
# ------------------------------------------------------------------------------
|
|
# Codec Training Configuration (codec_train.py)
|
|
# ------------------------------------------------------------------------------
|
|
codec:
|
|
sample_rate: 32000
|
|
batch_size: 16
|
|
num_epochs: 100
|
|
learning_rate: 1.0e-4
|
|
freeze_encoder: false
|
|
# Encoder/Decoder architecture params
|
|
encoder_dim: 768
|
|
encoder_num_layers: 8
|
|
bottleneck_channels: 5
|
|
decoder_num_layers: 8
|
|
|
|
# ------------------------------------------------------------------------------
|
|
# LLM Training Configuration (train_llm.py)
|
|
# ------------------------------------------------------------------------------
|
|
llm:
|
|
from_scratch: false
|
|
batch_size: 64
|
|
max_steps: 150000
|
|
max_lr: 2.0e-5
|
|
min_lr_ratio: 0.3
|
|
warmup_ratio: 0.3
|
|
cooldown_ratio: 0.1
|
|
grad_accum_steps: 1
|
|
seq_len: 1024
|
|
val_freq: 250
|
|
save_freq: 5000
|
|
text_factor: 0.5
|
|
betas: [0.9, 0.95]
|
|
weight_decay: 0.1
|
|
|
|
# ------------------------------------------------------------------------------
|
|
# Decoder (Vocos) Training Configuration (train_decoder.py)
|
|
# ------------------------------------------------------------------------------
|
|
decoder:
|
|
use_discriminator: true
|
|
batch_size: 8 # Reduced to avoid OOM, adjust based on your GPU
|
|
max_steps: 200000
|
|
max_lr: 2.0e-4
|
|
min_lr_ratio: 0.1
|
|
warmup_ratio: 0.2
|
|
cooldown_ratio: 0.1
|
|
grad_accum_steps: 1
|
|
seq_len: 1024
|
|
segment_size_samples: 32768 # ~1 sec (16 tokens)
|
|
val_freq: 250
|
|
val_steps: 10
|
|
save_freq: 3000
|
|
text_factor: 0.0
|
|
betas: [0.8, 0.99]
|
|
weight_decay: 0.1
|
|
start_step: 0
|
|
|
|
# Loss Weights
|
|
lambda_mel: 45.0
|
|
lambda_fm: 2.0
|
|
lambda_gen: 1.0
|
|
lambda_stft: 1.0
|
|
|
|
# ------------------------------------------------------------------------------
|
|
# Inference Configuration (simple_inference.py)
|
|
# ------------------------------------------------------------------------------
|
|
inference:
|
|
temperature: 0.8
|
|
top_k: 50
|
|
top_p: 0.95
|
|
repetition_penalty: 1.2
|
|
max_new_tokens: 512 |