instruct-MusicGen / src /constants.py
Vansh Chugh
initial deploy
21e9df7
Raw
History Blame Contribute Delete
1.23 kB
"""Constants used throughout the InstructMusicGen project."""
# Audio processing constants
SAMPLE_RATE = 32000 # MusicGen uses 32kHz sample rate
DEFAULT_RESAMPLING_RATE = 16000 # Standard resampling rate for evaluation metrics
# Model constants
CODEBOOK_SIZE = 2048 # Size of the codebook used in compression model
MAX_SEQUENCE_LENGTH = 1500 # Maximum sequence length for generation
# Training constants
DEFAULT_WARMUP_STEPS = 100 # Number of warmup steps for learning rate scheduler
DEFAULT_TRAINING_STEPS = 10000 # Default total training steps
# Generation constants
DEFAULT_NUM_SAMPLES = 1 # Default number of samples to generate
MAX_GENERATION_SAMPLES = 10 # Maximum number of generation samples to log
# Data processing constants
DEFAULT_TIME_IN_SECONDS = 5 # Default audio duration in seconds
MONO_CHANNELS = 1 # Number of channels for mono audio
STEREO_CHANNELS = 2 # Number of channels for stereo audio
# Evaluation constants
EVAL_SAMPLE_RATE = 16000 # Sample rate for evaluation metrics
DEFAULT_MEL_BINS = 128 # Default number of mel bins for spectrogram
# Retry constants
MAX_RETRIES = 3 # Maximum number of retries for operations that may fail
RETRY_DELAY = 1.0 # Delay in seconds between retries