1 files changed, 34 insertions, 0 deletions
diff --git a/modules/sd_hijack_xlmr.py b/modules/sd_hijack_xlmr.py
new file mode 100644
index 00000000..4ac51c38
--- /dev/null
+++ b/modules/sd_hijack_xlmr.py
@@ -0,0 +1,34 @@
+import open_clip.tokenizer
+import torch
+
+from modules import sd_hijack_clip, devices
+from modules.shared import opts
+
+
+class FrozenXLMREmbedderWithCustomWords(sd_hijack_clip.FrozenCLIPEmbedderWithCustomWords):
+    def __init__(self, wrapped, hijack):
+        super().__init__(wrapped, hijack)
+
+        self.id_start = wrapped.config.bos_token_id
+        self.id_end = wrapped.config.eos_token_id
+        self.id_pad = wrapped.config.pad_token_id
+
+        self.comma_token = self.tokenizer.get_vocab().get(',', None)  # alt diffusion doesn't have </w> bits for comma
+
+    def encode_with_transformers(self, tokens):
+        # there's no CLIP Skip here because all hidden layers have size of 1024 and the last one uses a
+        # trained layer to transform those 1024 into 768 for unet; so you can't choose which transformer
+        # layer to work with - you have to use the last
+
+        attention_mask = (tokens != self.id_pad).to(device=tokens.device, dtype=torch.int64)
+        features = self.wrapped(input_ids=tokens, attention_mask=attention_mask)
+        z = features['projection_state']
+
+        return z
+
+    def encode_embedding_init_text(self, init_text, nvpt):
+        embedding_layer = self.wrapped.roberta.embeddings
+        ids = self.wrapped.tokenizer(init_text, max_length=nvpt, return_tensors="pt", add_special_tokens=False)["input_ids"]
+        embedded = embedding_layer.token_embedding.wrapped(ids.to(devices.device)).squeeze(0)
+
+        return embedded