(self, image_embeds: torch.Tensor)
| 1581 | self.norm = nn.LayerNorm(cross_attention_dim) |
| 1582 | |
| 1583 | def forward(self, image_embeds: torch.Tensor): |
| 1584 | batch_size = image_embeds.shape[0] |
| 1585 | |
| 1586 | # image |
| 1587 | image_embeds = self.image_embeds(image_embeds.to(self.image_embeds.weight.dtype)) |
| 1588 | image_embeds = image_embeds.reshape(batch_size, self.num_image_text_embeds, -1) |
| 1589 | image_embeds = self.norm(image_embeds) |
| 1590 | return image_embeds |
| 1591 | |
| 1592 | |
| 1593 | class IPAdapterFullImageProjection(nn.Module): |