Module quantized_moondream 

Implementation of a quantized Moondream vision language model.

Moondream is a lightweight vision-language model for image understanding and generation. This module provides a quantized version for reduced memory usage and faster inference.

Key features:

• ViT-based vision encoder
• Phi-2 text decoder model
• Memory efficient 8-bit quantization
• Optimized for efficient deployment

References:

• Moondream Model

Structs

Encoder

Model

VisionEncoder