Apply patch [skip ci]

libs/partners/huggingface/langchain_huggingface/embeddings/huggingface.py

@@ -61,40 +61,47 @@ class HuggingFaceEmbeddings(BaseModel, Embeddings):
     """Whether to show a progress bar."""
 
     def __init__(self, **kwargs: Any):
-        """Initialize the sentence_transformer."""
+        """Initialize the transformer model."""
         super().__init__(**kwargs)
         try:
-            import sentence_transformers  # type: ignore[import]
+            from transformers import AutoModel, AutoTokenizer
         except ImportError as exc:
             msg = (
-                "Could not import sentence_transformers python package. "
+                "Could not import transformers python package. "
-                "Please install it with `pip install sentence-transformers`."
+                "Please install it with `pip install transformers`."
             )
             raise ImportError(msg) from exc
 
-        if self.model_kwargs.get("backend", "torch") == "ipex":
+        # Extract device from model_kwargs
-            if not is_optimum_intel_available() or not is_ipex_available():
+        self.device = self.model_kwargs.get("device", "cpu")
-                msg = f"Backend: ipex {IMPORT_ERROR.format('optimum[ipex]')}"
+        if isinstance(self.device, str):
-                raise ImportError(msg)
+            self.device = torch.device(self.device)
         
-            if is_optimum_intel_version("<", _MIN_OPTIMUM_VERSION):
+        # Remove device from model_kwargs as it's not a valid argument for from_pretrained
-                msg = (
+        model_kwargs = {k: v for k, v in self.model_kwargs.items() if k != "device"}
-                    f"Backend: ipex requires optimum-intel>="
+        
-                    f"{_MIN_OPTIMUM_VERSION}. You can install it with pip: "
+        # Load tokenizer and model
-                    "`pip install --upgrade --upgrade-strategy eager "
+        self.tokenizer = AutoTokenizer.from_pretrained(
-                    "`optimum[ipex]`."
+            self.model_name, 
+            cache_dir=self.cache_folder,
+            **model_kwargs
         )
-                raise ImportError(msg)
+        self.model = AutoModel.from_pretrained(
+            self.model_name,
+            cache_dir=self.cache_folder,
+            **model_kwargs
+        )
+        self.model.to(self.device)
+        self.model.eval()
         
-            from optimum.intel import IPEXSentenceTransformer  # type: ignore[import]
+        # Warn about multi-process not being supported
-
+        if self.multi_process:
-            model_cls = IPEXSentenceTransformer
+            import warnings
-
+            warnings.warn(
-        else:
+                "Multi-process encoding is not supported with the transformers "
-            model_cls = sentence_transformers.SentenceTransformer
+                "implementation. This parameter will be ignored.",
-
+                UserWarning,
-        self._client = model_cls(
+                stacklevel=2
-            self.model_name, cache_folder=self.cache_folder, **self.model_kwargs
             )
 
     model_config = ConfigDict(
@@ -171,3 +178,4 @@ class HuggingFaceEmbeddings(BaseModel, Embeddings):
         return self._embed([text], embed_kwargs)[0]
 
 
+