add embed function for voyageai multimodal embedder (#393)

* add embed function for voyageai multimodal embedder

* update changelog and version

* fix mixedbread ai embed unit test

Author: bryan-unstructured

CHANGELOG.md

@@ -1,3 +1,9 @@
+## 0.5.6-dev1
+
+### Fixes
+
+* **Fix voyageai embedder: add multimodal embedder function**
+
 ## 0.5.6
 
 ### Enhancements

test/integration/embedders/test_mixedbread.py

@@ -50,7 +50,7 @@ def test_raw_mixedbread_embedder(embedder_file: Path):
         embedder=embedder,
         embedder_file=embedder_file,
         expected_dimension=1024,
-        expected_is_unit_vector=False,
+        expected_is_unit_vector=True,
     )
 
 
@@ -67,5 +67,5 @@ async def test_raw_async_mixedbread_embedder(embedder_file: Path):
         embedder=embedder,
         embedder_file=embedder_file,
         expected_dimension=1024,
-        expected_is_unit_vector=False,
+        expected_is_unit_vector=True,
     )

test/integration/embedders/test_voyageai.py

@@ -61,3 +61,19 @@ async def test_raw_async_voyageai_embedder(embedder_file: Path):
     await validate_raw_embedder_async(
         embedder=embedder, embedder_file=embedder_file, expected_dimension=1024
     )
+
+
+@requires_env(API_KEY)
+def test_voyageai_multimodal_embedder(embedder_file: Path):
+    api_key = get_api_key()
+    embedder_config = EmbedderConfig(
+        embedding_provider="voyageai",
+        embedding_api_key=api_key,
+        embedding_model_name="voyage-multimodal-3",
+    )
+    embedder = Embedder(config=embedder_config)
+    results = embedder.run(elements_filepath=embedder_file)
+    assert results
+    with embedder_file.open("r") as f:
+        original_elements = json.load(f)
+    validate_embedding_output(original_elements=original_elements, output_elements=results)

unstructured_ingest/__version__.py

@@ -1 +1 @@
-__version__ = "0.5.6"  # pragma: no cover
+__version__ = "0.5.6-dev1"  # pragma: no cover

unstructured_ingest/embed/interfaces.py

@@ -49,7 +49,7 @@ def get_exemplary_embedding(self) -> list[float]:
     def is_unit_vector(self) -> bool:
         """Denotes if the embedding vector is a unit vector."""
         exemplary_embedding = self.get_exemplary_embedding()
-        return np.isclose(np.linalg.norm(exemplary_embedding), 1.0)
+        return np.isclose(np.linalg.norm(exemplary_embedding), 1.0, rtol=1e-03)
 
     def get_client(self):
         raise NotImplementedError
@@ -103,7 +103,7 @@ async def get_exemplary_embedding(self) -> list[float]:
     async def is_unit_vector(self) -> bool:
         """Denotes if the embedding vector is a unit vector."""
         exemplary_embedding = await self.get_exemplary_embedding()
-        return np.isclose(np.linalg.norm(exemplary_embedding), 1.0)
+        return np.isclose(np.linalg.norm(exemplary_embedding), 1.0, rtol=1e-03)
 
     def get_client(self):
         raise NotImplementedError

unstructured_ingest/embed/voyageai.py

@@ -96,7 +96,11 @@ def get_client(self) -> "VoyageAIClient":
         return self.config.get_client()
 
     def embed_batch(self, client: "VoyageAIClient", batch: list[str]) -> list[list[float]]:
-        response = client.embed(texts=batch, model=self.config.embedder_model_name)
+        if self.config.embedder_model_name == "voyage-multimodal-3":
+            batch = [[text] for text in batch]
+            response = client.multimodal_embed(inputs=batch, model=self.config.embedder_model_name)
+        else:
+            response = client.embed(texts=batch, model=self.config.embedder_model_name)
         return response.embeddings
 
 
@@ -113,5 +117,11 @@ def get_client(self) -> "AsyncVoyageAIClient":
     async def embed_batch(
         self, client: "AsyncVoyageAIClient", batch: list[str]
     ) -> list[list[float]]:
-        response = await client.embed(texts=batch, model=self.config.embedder_model_name)
+        if self.config.embedder_model_name == "voyage-multimodal-3":
+            batch = [[text] for text in batch]
+            response = await client.multimodal_embed(
+                inputs=batch, model=self.config.embedder_model_name
+            )
+        else:
+            response = await client.embed(texts=batch, model=self.config.embedder_model_name)
         return response.embeddings