Potter/mixedbread embedder (#3513)

Thanks to @huangrpablo and @juliuslipp we now have a mixedbread.ai
embedder!

Author: potter-potter

Diff for: .github/workflows/ci.yml

@@ -362,6 +362,7 @@ jobs:
         PINECONE_API_KEY: ${{secrets.PINECONE_API_KEY}}
         ASTRA_DB_APPLICATION_TOKEN: ${{secrets.ASTRA_DB_TOKEN}}
         ASTRA_DB_API_ENDPOINT: ${{secrets.ASTRA_DB_ENDPOINT}}
+        MXBAI_API_KEY: ${{secrets.MXBAI_API_KEY}}
         OCR_AGENT: "unstructured.partition.utils.ocr_models.tesseract_ocr.OCRAgentTesseract"
         CI: "true"
       run: |

Diff for: CHANGELOG.md

@@ -6,6 +6,8 @@
 
 ### Features
 
+* **Add MixedbreadAI embedder** Adds MixedbreadAI embeddings to support embedding via Mixedbread AI.
+
 ### Fixes
 
 * **Minify text_as_html from DOCX.** Previously `.metadata.text_as_html` for DOCX tables was "bloated" with whitespace and noise elements introduced by `tabulate` that produced over-chunking and lower "semantic density" of elements. Reduce HTML to minimum character count without preserving all text.

Diff for: MANIFEST.in

@@ -31,6 +31,7 @@ include requirements/ingest/dropbox.in
 include requirements/ingest/elasticsearch.in
 include requirements/ingest/embed-aws-bedrock.in
 include requirements/ingest/embed-huggingface.in
+include requirements/ingest/embed-mixedbreadai.in
 include requirements/ingest/embed-openai.in
 include requirements/ingest/gcs.in
 include requirements/ingest/github.in

Diff for: requirements/ingest/embed-mixedbreadai.in

@@ -0,0 +1,3 @@
+-c ../deps/constraints.txt
+-c ../base.txt
+mixedbread-ai

Diff for: requirements/ingest/embed-mixedbreadai.txt

@@ -0,0 +1,57 @@
+#
+# This file is autogenerated by pip-compile with Python 3.9
+# by the following command:
+#
+#    pip-compile ./ingest/embed-mixedbreadai.in
+#
+annotated-types==0.7.0
+    # via pydantic
+anyio==4.4.0
+    # via
+    #   -c ./ingest/../base.txt
+    #   httpx
+certifi==2024.7.4
+    # via
+    #   -c ./ingest/../base.txt
+    #   -c ./ingest/../deps/constraints.txt
+    #   httpcore
+    #   httpx
+exceptiongroup==1.2.2
+    # via
+    #   -c ./ingest/../base.txt
+    #   anyio
+h11==0.14.0
+    # via
+    #   -c ./ingest/../base.txt
+    #   httpcore
+httpcore==1.0.5
+    # via
+    #   -c ./ingest/../base.txt
+    #   httpx
+httpx==0.27.0
+    # via
+    #   -c ./ingest/../base.txt
+    #   mixedbread-ai
+idna==3.8
+    # via
+    #   -c ./ingest/../base.txt
+    #   anyio
+    #   httpx
+mixedbread-ai==2.2.6
+    # via -r ./ingest/embed-mixedbreadai.in
+pydantic==2.8.2
+    # via mixedbread-ai
+pydantic-core==2.20.1
+    # via pydantic
+sniffio==1.3.1
+    # via
+    #   -c ./ingest/../base.txt
+    #   anyio
+    #   httpx
+typing-extensions==4.12.2
+    # via
+    #   -c ./ingest/../base.txt
+    #   anyio
+    #   mixedbread-ai
+    #   pydantic
+    #   pydantic-core

Diff for: setup.py

@@ -171,6 +171,7 @@ def load_requirements(file_list: Optional[Union[str, List[str]]] = None) -> List
         "local-inference": all_doc_reqs,
         "paddleocr": load_requirements("requirements/extra-paddleocr.in"),
         "embed-huggingface": load_requirements("requirements/ingest/embed-huggingface.in"),
+        "embed-mixedbreadai": load_requirements("requirements/ingest/embed-mixedbreadai.in"),
         "embed-octoai": load_requirements("requirements/ingest/embed-octoai.in"),
         "embed-vertexai": load_requirements("requirements/ingest/embed-vertexai.in"),
         "embed-voyageai": load_requirements("requirements/ingest/embed-voyageai.in"),

Diff for: test_unstructured/embed/test_mixedbreadai.py

@@ -0,0 +1,41 @@
+from unstructured.documents.elements import Text
+from unstructured.embed.mixedbreadai import (
+    MixedbreadAIEmbeddingConfig,
+    MixedbreadAIEmbeddingEncoder,
+)
+
+
+def test_embed_documents_does_not_break_element_to_dict(mocker):
+    mock_client = mocker.MagicMock()
+
+    def mock_embeddings(
+        model,
+        normalized,
+        encoding_format,
+        truncation_strategy,
+        request_options,
+        input,
+    ):
+        mock_response = mocker.MagicMock()
+        mock_response.data = [mocker.MagicMock(embedding=[i, i + 1]) for i in range(len(input))]
+        return mock_response
+
+    mock_client.embeddings.side_effect = mock_embeddings
+
+    # Mock create_client to return our mock_client
+    mocker.patch.object(MixedbreadAIEmbeddingEncoder, "create_client", return_value=mock_client)
+
+    encoder = MixedbreadAIEmbeddingEncoder(
+        config=MixedbreadAIEmbeddingConfig(
+            api_key="api_key", model_name="mixedbread-ai/mxbai-embed-large-v1"
+        )
+    )
+
+    elements = encoder.embed_documents(
+        elements=[Text("This is sentence 1"), Text("This is sentence 2")],
+    )
+    assert len(elements) == 2
+    assert elements[0].to_dict()["text"] == "This is sentence 1"
+    assert elements[1].to_dict()["text"] == "This is sentence 2"
+    assert elements[0].embeddings is not None
+    assert elements[1].embeddings is not None