Module quantized_blip

Module quantized_blip 

Source
Expand description

BLIP model implementation with quantization support.

BLIP is a vision-language model for image understanding and generation tasks. This implementation provides quantization for reduced memory and compute.

Key characteristics:

  • Vision encoder using ViT architecture
  • Text decoder using BERT-style transformer
  • Cross-attention between vision and text features
  • Support for 8-bit quantization

References:

Re-exports§

pub use crate::quantized_var_builder::VarBuilder;

Structs§

BlipForConditionalGeneration
VisionModel

Type Aliases§

Config
VisionConfig