Add clip skip option to prompt node

invokeai/backend/model_management/lora.py

@@ -615,6 +615,24 @@ class ModelPatcher:
                 text_encoder.resize_token_embeddings(init_tokens_count)
 
 
+    @classmethod
+    @contextmanager
+    def apply_clip_skip(
+        cls,
+        text_encoder: CLIPTextModel,
+        clip_skip: int,
+    ):
+        skipped_layers = []
+        try:
+            for i in range(clip_skip):
+                skipped_layers.append(text_encoder.text_model.encoder.layers.pop(-1))
+
+            yield
+
+        finally:
+            while len(skipped_layers) > 0:
+                text_encoder.text_model.encoder.layers.append(skipped_layers.pop())
+
 class TextualInversionModel:
     name: str
     embedding: torch.Tensor # [n, 768]|[n, 1280]