Refactoring VoyageAI integration (#3878)

Using VoyageAI's python package directly, allowing more features than
through langchain

Author: fzowl

Diff for: CHANGELOG.md

@@ -1,6 +1,7 @@
 ## 0.16.17-dev1
 
 ### Enhancements
+- **Refactoring the VoyageAI integration** to use voyageai package directly, allowing extra features.
 
 ### Features
 

Diff for: test_unstructured/embed/test_voyageai.py

@@ -1,17 +1,21 @@
+from unittest.mock import Mock
+
 from unstructured.documents.elements import Text
 from unstructured.embed.voyageai import VoyageAIEmbeddingConfig, VoyageAIEmbeddingEncoder
 
 
 def test_embed_documents_does_not_break_element_to_dict(mocker):
     # Mocked client with the desired behavior for embed_documents
+    embed_response = Mock()
+    embed_response.embeddings = [[1], [2]]
     mock_client = mocker.MagicMock()
-    mock_client.embed_documents.return_value = [1, 2]
+    mock_client.embed.return_value = embed_response
 
     # Mock get_client to return our mock_client
     mocker.patch.object(VoyageAIEmbeddingConfig, "get_client", return_value=mock_client)
 
     encoder = VoyageAIEmbeddingEncoder(
-        config=VoyageAIEmbeddingConfig(api_key="api_key", model_name="voyage-law-2")
+        config=VoyageAIEmbeddingConfig(api_key="api_key", model_name="voyage-3-large")
     )
     elements = encoder.embed_documents(
         elements=[Text("This is sentence 1"), Text("This is sentence 2")],

Diff for: test_unstructured_ingest/src/local-embed-voyageai.sh

@@ -37,7 +37,7 @@ PYTHONPATH=${PYTHONPATH:-.} "$RUN_SCRIPT" \
   --work-dir "$WORK_DIR" \
   --embedding-provider "voyageai" \
   --embedding-api-key "$VOYAGE_API_KEY" \
-  --embedding-model-name "voyage-large-2"
+  --embedding-model-name "voyage-3-large"
 
 set +e
 

Diff for: unstructured/embed/__init__.py

@@ -13,7 +13,7 @@
     "langchain-huggingface": HuggingFaceEmbeddingEncoder,
     "langchain-aws-bedrock": BedrockEmbeddingEncoder,
     "langchain-vertexai": VertexAIEmbeddingEncoder,
-    "langchain-voyageai": VoyageAIEmbeddingEncoder,
+    "voyageai": VoyageAIEmbeddingEncoder,
     "mixedbread-ai": MixedbreadAIEmbeddingEncoder,
     "octoai": OctoAIEmbeddingEncoder,
 }

Diff for: unstructured/embed/voyageai.py

@@ -1,5 +1,5 @@
 from dataclasses import dataclass
-from typing import TYPE_CHECKING, List, Optional
+from typing import TYPE_CHECKING, Iterable, List, Optional, cast
 
 import numpy as np
 from pydantic import Field, SecretStr
@@ -9,30 +9,46 @@
 from unstructured.utils import requires_dependencies
 
 if TYPE_CHECKING:
-    from langchain_voyageai import VoyageAIEmbeddings
+    from voyageai import Client
+
+DEFAULT_VOYAGE_2_BATCH_SIZE = 72
+DEFAULT_VOYAGE_3_LITE_BATCH_SIZE = 30
+DEFAULT_VOYAGE_3_BATCH_SIZE = 10
+DEFAULT_BATCH_SIZE = 7
 
 
 class VoyageAIEmbeddingConfig(EmbeddingConfig):
     api_key: SecretStr
     model_name: str
+    show_progress_bar: bool = False
     batch_size: Optional[int] = Field(default=None)
     truncation: Optional[bool] = Field(default=None)
+    output_dimension: Optional[int] = Field(default=None)
 
     @requires_dependencies(
-        ["langchain", "langchain_voyageai"],
+        ["voyageai"],
         extras="embed-voyageai",
     )
-    def get_client(self) -> "VoyageAIEmbeddings":
-        """Creates a Langchain VoyageAI python client to embed elements."""
-        from langchain_voyageai import VoyageAIEmbeddings
-
-        return VoyageAIEmbeddings(
-            voyage_api_key=self.api_key,
-            model=self.model_name,
-            batch_size=self.batch_size,
-            truncation=self.truncation,
+    def get_client(self) -> "Client":
+        """Creates a VoyageAI python client to embed elements."""
+        from voyageai import Client
+
+        return Client(
+            api_key=self.api_key.get_secret_value(),
         )
 
+    def get_batch_size(self):
+        if self.batch_size is None:
+            if self.model_name in ["voyage-2", "voyage-02"]:
+                self.batch_size = DEFAULT_VOYAGE_2_BATCH_SIZE
+            elif self.model_name == "voyage-3-lite":
+                self.batch_size = DEFAULT_VOYAGE_3_LITE_BATCH_SIZE
+            elif self.model_name == "voyage-3":
+                self.batch_size = DEFAULT_VOYAGE_3_BATCH_SIZE
+            else:
+                self.batch_size = DEFAULT_BATCH_SIZE
+        return self.batch_size
+
 
 @dataclass
 class VoyageAIEmbeddingEncoder(BaseEmbeddingEncoder):
@@ -56,12 +72,29 @@ def is_unit_vector(self) -> bool:
 
     def embed_documents(self, elements: List[Element]) -> List[Element]:
         client = self.config.get_client()
-        embeddings = client.embed_documents([str(e) for e in elements])
+        embeddings: List[List[float]] = []
+
+        _iter = self._get_batch_iterator(elements)
+        for i in _iter:
+            r = client.embed(
+                texts=[str(e) for e in elements[i : i + self.config.get_batch_size()]],
+                model=self.config.model_name,
+                input_type="document",
+                truncation=self.config.truncation,
+                output_dimension=self.config.output_dimension,
+            ).embeddings
+            embeddings.extend(cast(Iterable[List[float]], r))
         return self._add_embeddings_to_elements(elements, embeddings)
 
     def embed_query(self, query: str) -> List[float]:
         client = self.config.get_client()
-        return client.embed_query(query)
+        return client.embed(
+            texts=[query],
+            model=self.config.model_name,
+            input_type="query",
+            truncation=self.config.truncation,
+            output_dimension=self.config.output_dimension,
+        ).embeddings[0]
 
     @staticmethod
     def _add_embeddings_to_elements(elements, embeddings) -> List[Element]:
@@ -71,3 +104,19 @@ def _add_embeddings_to_elements(elements, embeddings) -> List[Element]:
             element.embeddings = embeddings[i]
             elements_w_embedding.append(element)
         return elements
+
+    def _get_batch_iterator(self, elements: List[Element]) -> Iterable:
+        if self.config.show_progress_bar:
+            try:
+                from tqdm.auto import tqdm  # type: ignore
+            except ImportError as e:
+                raise ImportError(
+                    "Must have tqdm installed if `show_progress_bar` is set to True. "
+                    "Please install with `pip install tqdm`."
+                ) from e
+
+            _iter = tqdm(range(0, len(elements), self.config.get_batch_size()))
+        else:
+            _iter = range(0, len(elements), self.config.get_batch_size())  # type: ignore
+
+        return _iter