Create note_these_are_token_vectors.txt
Browse files
vocab/token_vectors/note_these_are_token_vectors.txt
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Although they have the dimension of 1x768 , the token_vectors are **not** text_encodings.
|
2 |
+
|
3 |
+
The token vectors , being vectors of constant value that represent an item in the vocab.json file , are stored within the SD / FLUX models
|
4 |
+
|
5 |
+
, as the "building blocks" for CLIP , from which the text_encoding is produced.
|
6 |
+
|
7 |
+
Do not bother comparing token vectors with text_encodings. You will get meaningless results.
|
8 |
+
|
9 |
+
Keep token_vectors and text_encodings separate.
|