Add SigLIP support

sentence_transformers/SentenceTransformer.py

@@ -1198,8 +1198,17 @@ def _load_auto_model(
         local_files_only: bool = False,
     ):
         """
-        Creates a simple Transformer + Mean Pooling model and returns the modules
+        Creates a simple Transformer + Mean Pooling model or a SigLIP model and returns the modules
         """
+        if "siglip" in model_name_or_path:
+            from sentence_transformers.models import SigLIPModel
+            logger.warning(
+                "No sentence-transformers model found with name {}. Creating a SigLIPModel.".format(
+                    model_name_or_path
+                )
+            )
+            return [SigLIPModel(model_name_or_path)]
+
         logger.warning(
             "No sentence-transformers model found with name {}. Creating a new one with MEAN pooling.".format(
                 model_name_or_path

sentence_transformers/models/SigLIPModel.py

@@ -0,0 +1,81 @@
+from typing import Union
+from torch import nn
+import transformers
+import torch
+from PIL import Image
+
+
+class SigLIPModel(nn.Module):
+    def __init__(self, model_name: str = "google/siglip-so400m-patch14-384", processor_name=None):
+        super(SigLIPModel, self).__init__()
+
+        if processor_name is None:
+            processor_name = model_name
+
+        self.model = transformers.AutoModel.from_pretrained(model_name)
+        self.processor = transformers.AutoProcessor.from_pretrained(processor_name)
+
+    def __repr__(self):
+        return "SigLIPModel()"
+
+    def forward(self, features):
+        image_embeds = []
+        text_embeds = []
+
+        if "pixel_values" in features:
+            image_embeds = self.model.get_image_features(features["pixel_values"])
+
+        if "input_ids" in features:
+            text_embeds = self.model.get_text_features(
+                input_ids=features.get("input_ids"),
+                attention_mask=features.get("attention_mask", None),
+                position_ids=features.get("position_ids", None),
+                output_attentions=features.get("output_attentions", None),
+                output_hidden_states=features.get("output_hidden_states", None),
+            )
+
+        sentence_embedding = []
+        image_features = iter(image_embeds)
+        text_features = iter(text_embeds)
+
+        for idx, input_type in enumerate(features["image_text_info"]):
+            if input_type == 0:
+                sentence_embedding.append(next(image_features))
+            else:
+                sentence_embedding.append(next(text_features))
+
+        features["sentence_embedding"] = torch.stack(sentence_embedding).float()
+
+        return features
+
+    def tokenize(self, texts, padding: Union[str, bool] = "max_length"):
+        images = []
+        texts_values = []
+        image_text_info = []
+
+        for idx, data in enumerate(texts):
+            if isinstance(data, Image.Image):  # An Image
+                images.append(data)
+                image_text_info.append(0)
+            else:  # A text
+                texts_values.append(data)
+                image_text_info.append(1)
+
+        encoding = {}
+        if len(texts_values):
+            encoding = self.processor.tokenizer(texts_values, return_tensors="pt", padding=padding)
+
+        if len(images):
+            image_features = self.processor.image_processor(images, return_tensors="pt")
+            encoding["pixel_values"] = image_features.pixel_values
+
+        encoding["image_text_info"] = image_text_info
+        return encoding
+
+    def save(self, output_path: str):
+        self.model.save_pretrained(output_path)
+        self.processor.save_pretrained(output_path)
+
+    @staticmethod
+    def load(input_path: str):
+        return SigLIPModel(model_name=input_path)

sentence_transformers/models/__init__.py

@@ -12,6 +12,7 @@
 from .WordEmbeddings import WordEmbeddings
 from .WordWeights import WordWeights
 from .CLIPModel import CLIPModel
+from .SigLIPModel import SigLIPModel
 
 __all__ = [
     "Transformer",
@@ -28,4 +29,5 @@
     "WordEmbeddings",
     "WordWeights",
     "CLIPModel",
+    "SigLIPModel"
 ]