Expand description
Joint Embedding Spaces for Cross-Modal Vector Search
This module implements advanced joint embedding spaces that enable:
- CLIP-style text-image alignment
- Cross-modal attention mechanisms
- Contrastive learning for alignment
- Multi-modal fusion strategies
- Domain adaptation and transfer learning
Structs§
- Alignment
Pair - Alignment pair for caching and training
- CLIP
Aligner - CLIP-style contrastive learning implementation
- Contrastive
Optimizer - Contrastive optimizer for alignment training
- Cross
Modal Attention - Cross-modal attention mechanism for joint spaces
- Curriculum
Learning - Curriculum learning for progressive training
- Data
Augmentation - Data augmentation for improved generalization
- Domain
Adapter - Domain adaptation module for cross-domain alignment
- Domain
Classifier - Domain
Statistics - Joint
Embedding Config - Configuration for joint embedding space
- Joint
Embedding Space - Joint embedding space for cross-modal alignment
- Linear
Projector - Linear projector for transforming embeddings to joint space
- Temperature
Scheduler - Temperature scheduler for contrastive learning
- Training
Statistics - Training statistics for monitoring
Enums§
- Activation
Function - Activation functions for projectors
- Audio
Augmentation - Difficulty
Schedule - Image
Augmentation - Learning
Rate Schedule - Pacing
Function - Schedule
Type - Text
Augmentation