Enum MusicGenModelSize

Namespace

AiDotNet.Audio.MusicGen

Assembly

AiDotNet.dll

Specifies the size variant of the MusicGen model.

public enum MusicGenModelSize

Fields

Large = 2

Large model variant (3.3B parameters).

• Text encoder: 1024 hidden dim
• LM: 2048 hidden dim, 48 layers, 16 heads
• Highest quality, requires significant GPU memory

Medium = 1

Medium model variant (1.5B parameters). Default choice.

• Text encoder: 768 hidden dim
• LM: 1536 hidden dim, 24 layers, 16 heads
• Good balance of quality and speed

Melody = 3

Melody model variant (1.5B parameters).

• Same architecture as Medium
• Additionally conditioned on melody input
• Can generate music that follows a given melody

Small = 0

Small model variant (300M parameters).

• Text encoder: 256 hidden dim
• LM: 1024 hidden dim, 24 layers, 16 heads
• Fast inference, suitable for real-time applications

Stereo = 4

Stereo model variant (1.5B parameters).

• Same architecture as Medium
• Generates stereo audio output
• Uses additional codebook for left/right channel separation

Remarks

MusicGen comes in different sizes balancing quality and computational requirements. Larger models produce higher quality music but require more memory and compute.

For Beginners: Think of model sizes like different quality levels:

• Small: Fast generation, good for experimentation (300M parameters)
• Medium: Balanced quality and speed (1.5B parameters)
• Large: Best quality, requires more resources (3.3B parameters)

Start with Small for testing, use Large for final production.