jinaai
/

jina-clip-implementation

🇪🇺 Region: EU

Model card Files Files and versions

Fix docs

#8

by Xenova HF Staff - opened Nov 26, 2024

base: refs/heads/main

←

from: refs/pr/8

Discussion Files changed

Files changed (1) hide show

modeling_clip.py +2 -2

modeling_clip.py CHANGED Viewed

@@ -410,7 +410,7 @@ class JinaCLIPModel(JinaCLIPPreTrainedModel):
                 from convert_to_numpy
             device(`torch.device`, *optional*, defaults to None):
                 Which torch.device to use for the computation
-            normalize_embeddings(`bool`, *optional*, defaults to False):
                 If set to true, returned vectors will have length 1. In that case,
                 the faster dot-product (util.dot_score) instead of cosine similarity
                 can be used
@@ -545,7 +545,7 @@ class JinaCLIPModel(JinaCLIPPreTrainedModel):
                 from convert_to_numpy
             device(`torch.device`, *optional*, defaults to None):
                 Which torch.device to use for the computation
-            normalize_embeddings(`bool`, *optional*, defaults to False):
                 If set to true, returned vectors will have length 1. In that case,
                 the faster dot-product (util.dot_score) instead of cosine similarity
                 can be used

                 from convert_to_numpy
             device(`torch.device`, *optional*, defaults to None):
                 Which torch.device to use for the computation
+            normalize_embeddings(`bool`, *optional*, defaults to True):
                 If set to true, returned vectors will have length 1. In that case,
                 the faster dot-product (util.dot_score) instead of cosine similarity
                 can be used
                 from convert_to_numpy
             device(`torch.device`, *optional*, defaults to None):
                 Which torch.device to use for the computation
+            normalize_embeddings(`bool`, *optional*, defaults to True):
                 If set to true, returned vectors will have length 1. In that case,
                 the faster dot-product (util.dot_score) instead of cosine similarity
                 can be used