Reduce half of similarity muls after encoding (openai#140)

(cAB)^T = c B^T A^T
Saves half of the similarity products in the CLIP model.py after the visual/text encoding stages

clip/model.py

@@ -362,7 +362,7 @@ def forward(self, image, text):
         # cosine similarity as logits
         logit_scale = self.logit_scale.exp()
         logits_per_image = logit_scale * image_features @ text_features.t()
-        logits_per_text = logit_scale * text_features @ image_features.t()
+        logits_per_text = logits_per_image.t()
 
         # shape = [global_batch_size, global_batch_size]
         return logits_per_image, logits_per_text