From da464a3fb39ecc6ea7b22fe87271194480d8501c Mon Sep 17 00:00:00 2001
From: AUTOMATIC1111 <16777216c@gmail.com>
Date: Wed, 12 Jul 2023 23:52:43 +0300
Subject: SDXL support

---
 modules/sd_hijack_open_clip.py | 38 ++++++++++++++++++++++++++++++++++----
 1 file changed, 34 insertions(+), 4 deletions(-)

(limited to 'modules/sd_hijack_open_clip.py')

diff --git a/modules/sd_hijack_open_clip.py b/modules/sd_hijack_open_clip.py
index 6ac5bda6..fcf5ad07 100644
--- a/modules/sd_hijack_open_clip.py
+++ b/modules/sd_hijack_open_clip.py
@@ -16,10 +16,6 @@ class FrozenOpenCLIPEmbedderWithCustomWords(sd_hijack_clip.FrozenCLIPEmbedderWit
         self.id_end = tokenizer.encoder["<end_of_text>"]
         self.id_pad = 0
 
-        self.is_trainable = getattr(wrapped, 'is_trainable', False)
-        self.input_key = getattr(wrapped, 'input_key', 'txt')
-        self.legacy_ucg_val = None
-
     def tokenize(self, texts):
         assert not opts.use_old_emphasis_implementation, 'Old emphasis implementation not supported for Open Clip'
 
@@ -39,3 +35,37 @@ class FrozenOpenCLIPEmbedderWithCustomWords(sd_hijack_clip.FrozenCLIPEmbedderWit
         embedded = self.wrapped.model.token_embedding.wrapped(ids).squeeze(0)
 
         return embedded
+
+
+class FrozenOpenCLIPEmbedder2WithCustomWords(sd_hijack_clip.FrozenCLIPEmbedderWithCustomWordsBase):
+    def __init__(self, wrapped, hijack):
+        super().__init__(wrapped, hijack)
+
+        self.comma_token = [v for k, v in tokenizer.encoder.items() if k == ',</w>'][0]
+        self.id_start = tokenizer.encoder["<start_of_text>"]
+        self.id_end = tokenizer.encoder["<end_of_text>"]
+        self.id_pad = 0
+
+    def tokenize(self, texts):
+        assert not opts.use_old_emphasis_implementation, 'Old emphasis implementation not supported for Open Clip'
+
+        tokenized = [tokenizer.encode(text) for text in texts]
+
+        return tokenized
+
+    def encode_with_transformers(self, tokens):
+        d = self.wrapped.encode_with_transformer(tokens)
+        z = d[self.wrapped.layer]
+
+        pooled = d.get("pooled")
+        if pooled is not None:
+            z.pooled = pooled
+
+        return z
+
+    def encode_embedding_init_text(self, init_text, nvpt):
+        ids = tokenizer.encode(init_text)
+        ids = torch.asarray([ids], device=devices.device, dtype=torch.int)
+        embedded = self.wrapped.model.token_embedding.wrapped(ids).squeeze(0)
+
+        return embedded
-- 
cgit v1.2.3