GenerativeArchitectureType
Enums · AiDotNet.VisionLanguage.Encoders
Specifies the architecture type for generative vision-language models.
Fields
| Field | Summary |
|---|---|
GenerativeArchitectureType CausalMultimodal |
Causal multimodal: visual tokens embedded directly in causal language model (KOSMOS). |
GenerativeArchitectureType EncoderDecoder |
Encoder-decoder: ViT encoder + autoregressive text decoder with cross-attention (GIT, CoCa, PaLI). |
GenerativeArchitectureType PerceiverResampler |
Perceiver resampler: latent queries cross-attend to vision, gated cross-attention into LLM (Flamingo, IDEFICS). |
GenerativeArchitectureType QFormerBridge |
Q-Former bridge: learnable queries cross-attend to vision, then feed decoder (InstructBLIP, BLIP-3). |
GenerativeArchitectureType UnifiedGeneration |
Unified generation: single model for both understanding and image/text generation (Emu). |
int value__ |