Fix CI and to_dict tests

.github/workflows/e2e_preview.yml

@@ -36,7 +36,7 @@ jobs:
         sudo apt install ffmpeg  # for local Whisper tests
 
     - name: Install Haystack
-      run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf tika 'azure-ai-formrecognizer>=3.2.0b2'
+      run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf tika 'azure-ai-formrecognizer>=3.2.0b2' gradientai
 
     - name: Run tests
       run: pytest e2e/preview

.github/workflows/linting_preview.yml

@@ -38,7 +38,7 @@ jobs:
           python-version: ${{ env.PYTHON_VERSION }}
 
       - name: Install Haystack
-        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf tika 'azure-ai-formrecognizer>=3.2.0b2'
+        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf tika 'azure-ai-formrecognizer>=3.2.0b2' gradientai
 
       - name: Mypy
         if: steps.files.outputs.any_changed == 'true'
@@ -69,7 +69,7 @@ jobs:
 
       - name: Install Haystack
         run: |
-          pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2'
+          pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2' gradientai
           pip install ./haystack-linter
 
       - name: Pylint

.github/workflows/tests_preview.yml

@@ -116,7 +116,7 @@ jobs:
           python-version: ${{ env.PYTHON_VERSION }}
 
       - name: Install Haystack
-        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2'
+        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2' gradientai
 
       - name: Run
         run: pytest -m "not integration" test/preview
@@ -174,7 +174,7 @@ jobs:
           sudo apt install ffmpeg  # for local Whisper tests
 
       - name: Install Haystack
-        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2'
+        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2' gradientai
 
       - name: Run
         run: pytest --maxfail=5 -m "integration" test/preview
@@ -230,7 +230,7 @@ jobs:
           colima start
 
       - name: Install Haystack
-        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2'
+        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2' gradientai
 
       - name: Run Tika
         run: docker run -d -p 9998:9998 apache/tika:2.9.0.0
@@ -281,7 +281,7 @@ jobs:
           python-version: ${{ env.PYTHON_VERSION }}
 
       - name: Install Haystack
-        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2'
+        run: pip install .[dev,preview,audio] langdetect transformers[torch,sentencepiece]==4.35.2 'sentence-transformers>=2.2.0' pypdf markdown-it-py mdit_plain tika 'azure-ai-formrecognizer>=3.2.0b2' gradientai
 
       - name: Run
         run: pytest --maxfail=5 -m "integration" test/preview -k 'not tika'

e2e/preview/pipelines/test_gradient_rag_pipelines.py

@@ -0,0 +1,92 @@
+import os
+import json
+import pytest
+
+from haystack.preview import Pipeline, Document
+from haystack.preview.components.embedders.gradient_document_embedder import GradientDocumentEmbedder
+from haystack.preview.components.embedders.gradient_text_embedder import GradientTextEmbedder
+from haystack.preview.document_stores import InMemoryDocumentStore
+from haystack.preview.components.writers import DocumentWriter
+from haystack.preview.components.retrievers import InMemoryEmbeddingRetriever
+from haystack.preview.components.generators.gradient.base import GradientGenerator
+from haystack.preview.components.builders.answer_builder import AnswerBuilder
+from haystack.preview.components.builders.prompt_builder import PromptBuilder
+
+
+@pytest.mark.skipif(
+    not os.environ.get("GRADIENT_ACCESS_TOKEN", None) or not os.environ.get("GRADIENT_WORKSPACE_ID", None),
+    reason="Export env variables called GRADIENT_ACCESS_TOKEN and GRADIENT_WORKSPACE_ID containing the Gradient configuration settings to run this test.",
+)
+def test_gradient_embedding_retrieval_rag_pipeline(tmp_path):
+    # Create the RAG pipeline
+    prompt_template = """
+    Given these documents, answer the question.\nDocuments:
+    {% for doc in documents %}
+        {{ doc.content }}
+    {% endfor %}
+
+    \nQuestion: {{question}}
+    \nAnswer:
+    """
+
+    gradient_access_token = os.environ.get("GRADIENT_ACCESS_TOKEN")
+    rag_pipeline = Pipeline()
+    embedder = GradientTextEmbedder(access_token=gradient_access_token)
+    rag_pipeline.add_component(instance=embedder, name="text_embedder")
+    rag_pipeline.add_component(
+        instance=InMemoryEmbeddingRetriever(document_store=InMemoryDocumentStore()), name="retriever"
+    )
+    rag_pipeline.add_component(instance=PromptBuilder(template=prompt_template), name="prompt_builder")
+    rag_pipeline.add_component(
+        instance=GradientGenerator(access_token=gradient_access_token, base_model_slug="llama2-7b-chat"), name="llm"
+    )
+    rag_pipeline.add_component(instance=AnswerBuilder(), name="answer_builder")
+    rag_pipeline.connect("text_embedder", "retriever")
+    rag_pipeline.connect("retriever", "prompt_builder.documents")
+    rag_pipeline.connect("prompt_builder", "llm")
+    rag_pipeline.connect("llm.replies", "answer_builder.replies")
+    rag_pipeline.connect("retriever", "answer_builder.documents")
+
+    # Draw the pipeline
+    rag_pipeline.draw(tmp_path / "test_gradient_embedding_rag_pipeline.png")
+
+    # Serialize the pipeline to JSON
+    with open(tmp_path / "test_bm25_rag_pipeline.json", "w") as f:
+        json.dump(rag_pipeline.to_dict(), f)
+
+    # Load the pipeline back
+    with open(tmp_path / "test_bm25_rag_pipeline.json", "r") as f:
+        rag_pipeline = Pipeline.from_dict(json.load(f))
+
+    # Populate the document store
+    documents = [
+        Document(content="My name is Jean and I live in Paris."),
+        Document(content="My name is Mark and I live in Berlin."),
+        Document(content="My name is Giorgio and I live in Rome."),
+    ]
+    document_store = rag_pipeline.get_component("retriever").document_store
+    indexing_pipeline = Pipeline()
+    indexing_pipeline.add_component(instance=GradientDocumentEmbedder(), name="document_embedder")
+    indexing_pipeline.add_component(instance=DocumentWriter(document_store=document_store), name="document_writer")
+    indexing_pipeline.connect("document_embedder", "document_writer")
+    indexing_pipeline.run({"document_embedder": {"documents": documents}})
+
+    # Query and assert
+    questions = ["Who lives in Paris?", "Who lives in Berlin?", "Who lives in Rome?"]
+    answers_spywords = ["Jean", "Mark", "Giorgio"]
+
+    for question, spyword in zip(questions, answers_spywords):
+        result = rag_pipeline.run(
+            {
+                "text_embedder": {"text": question},
+                "prompt_builder": {"question": question},
+                "answer_builder": {"query": question},
+            }
+        )
+
+        assert len(result["answer_builder"]["answers"]) == 1
+        generated_answer = result["answer_builder"]["answers"][0]
+        assert spyword in generated_answer.data
+        assert generated_answer.query == question
+        assert hasattr(generated_answer, "documents")
+        assert hasattr(generated_answer, "metadata")

haystack/preview/components/embedders/__init__.py

@@ -1,3 +1,6 @@
+from haystack.preview.components.embedders.gradient_text_embedder import GradientTextEmbedder
+from haystack.preview.components.embedders.gradient_document_embedder import GradientDocumentEmbedder
+
 from haystack.preview.components.embedders.sentence_transformers_text_embedder import SentenceTransformersTextEmbedder
 from haystack.preview.components.embedders.sentence_transformers_document_embedder import (
     SentenceTransformersDocumentEmbedder,
@@ -6,6 +9,8 @@
 from haystack.preview.components.embedders.openai_text_embedder import OpenAITextEmbedder
 
 __all__ = [
+    "GradientTextEmbedder",
+    "GradientDocumentEmbedder",
     "SentenceTransformersTextEmbedder",
     "SentenceTransformersDocumentEmbedder",
     "OpenAITextEmbedder",

haystack/preview/components/embedders/gradient_document_embedder.py

@@ -0,0 +1,112 @@
+import logging
+from typing import List, Optional, Dict, Any
+
+from haystack.preview import component, Document, default_to_dict
+from haystack.preview.lazy_imports import LazyImport
+
+with LazyImport(message="Run 'pip install gradientai'") as gradientai_import:
+    from gradientai import Gradient
+
+logger = logging.getLogger(__name__)
+
+
+@component
+class GradientDocumentEmbedder:
+    """
+    A component for computing Document embeddings using Gradient AI API..
+    The embedding of each Document is stored in the `embedding` field of the Document.
+
+    ```python
+    embedder = GradientDocumentEmbedder(
+        access_token=gradient_access_token,
+        workspace_id=gradient_workspace_id,
+        model_name="bge_large"))
+    p = Pipeline()
+    p.add_component(embedder, name="document_embedder")
+    p.add_component(instance=GradientDocumentEmbedder(
+    p.add_component(instance=DocumentWriter(document_store=InMemoryDocumentStore()), name="document_writer")
+    p.connect("document_embedder", "document_writer")
+    p.run({"document_embedder": {"documents": documents}})
+    ```
+    """
+
+    def __init__(
+        self,
+        *,
+        model_name: str = "bge-large",
+        batch_size: int = 100,
+        access_token: Optional[str] = None,
+        workspace_id: Optional[str] = None,
+        host: Optional[str] = None,
+    ) -> None:
+        """
+        Create a GradientDocumentEmbedder component.
+
+        :param model_name: The name of the model to use.
+        :param access_token: The Gradient access token. If not provided it's read from the environment
+                             variable GRADIENT_ACCESS_TOKEN.
+        :param workspace_id: The Gradient workspace ID. If not provided it's read from the environment
+                             variable GRADIENT_WORKSPACE_ID.
+        :param host: The Gradient host. By default it uses https://api.gradient.ai/.
+        """
+        gradientai_import.check()
+        self._batch_size = batch_size
+        self._host = host
+        self._model_name = model_name
+
+        self._gradient = Gradient(access_token=access_token, host=host, workspace_id=workspace_id)
+
+    def _get_telemetry_data(self) -> Dict[str, Any]:
+        """
+        Data that is sent to Posthog for usage analytics.
+        """
+        return {"model": self._model_name}
+
+    def to_dict(self) -> dict:
+        """
+        Serialize the component to a Python dictionary.
+        """
+        return default_to_dict(self, workspace_id=self._gradient.workspace_id, model_name=self._model_name)
+
+    def warm_up(self) -> None:
+        """
+        Load the embedding model.
+        """
+        if not hasattr(self, "_embedding_model"):
+            self._embedding_model = self._gradient.get_embeddings_model(slug=self._model_name)
+
+    def _generate_embeddings(self, documents: List[Document], batch_size: int) -> List[List[float]]:
+        """
+        Batches the documents and generates the embeddings.
+        """
+        batches = [documents[i : i + batch_size] for i in range(0, len(documents), batch_size)]
+
+        embeddings = []
+        for batch in batches:
+            response = self._embedding_model.generate_embeddings(inputs=[{"input": doc.content} for doc in batch])
+            embeddings.extend([e.embedding for e in response.embeddings])
+
+        return embeddings
+
+    @component.output_types(documents=List[Document])
+    def run(self, documents: List[Document]):
+        """
+        Embed a list of Documents.
+        The embedding of each Document is stored in the `embedding` field of the Document.
+
+        :param documents: A list of Documents to embed.
+        """
+        if not isinstance(documents, list) or documents and any(not isinstance(doc, Document) for doc in documents):
+            raise TypeError(
+                "GradientDocumentEmbedder expects a list of Documents as input."
+                "In case you want to embed a list of strings, please use the GradientTextEmbedder."
+            )
+
+        if not hasattr(self, "_embedding_model"):
+            raise RuntimeError("The embedding model has not been loaded. Please call warm_up() before running.")
+
+        embeddings = self._generate_embeddings(documents=documents, batch_size=self._batch_size)
+        for doc, embedding in zip(documents, embeddings):
+            doc.embedding = embedding
+
+        return {"documents": documents}

haystack/preview/components/embedders/gradient_text_embedder.py

@@ -0,0 +1,88 @@
+from typing import Any, Dict, List, Optional
+
+from haystack.preview import component, default_to_dict
+from haystack.preview.lazy_imports import LazyImport
+
+with LazyImport(message="Run 'pip install gradientai'") as gradientai_import:
+    from gradientai import Gradient
+
+
+@component
+class GradientTextEmbedder:
+    """
+    A component for embedding strings using models hosted on Gradient AI (https://gradient.ai).
+
+    ```python
+    embedder = GradientTextEmbedder(
+        access_token=gradient_access_token,
+        workspace_id=gradient_workspace_id,
+        model_name="bge_large")
+    p = Pipeline()
+    p.add_component(instance=embedder, name="text_embedder")
+    p.add_component(instance=InMemoryEmbeddingRetriever(document_store=InMemoryDocumentStore()), name="retriever")
+    p.connect("text_embedder", "retriever")
+    p.run("embed me!!!")
+    ```
+    """
+
+    def __init__(
+        self,
+        *,
+        model_name: str = "bge-large",
+        access_token: Optional[str] = None,
+        workspace_id: Optional[str] = None,
+        host: Optional[str] = None,
+    ) -> None:
+        """
+        Create a GradientTextEmbedder component.
+
+        :param model_name: The name of the model to use.
+        :param access_token: The Gradient access token. If not provided it's read from the environment
+                             variable GRADIENT_ACCESS_TOKEN.
+        :param workspace_id: The Gradient workspace ID. If not provided it's read from the environment
+                             variable GRADIENT_WORKSPACE_ID.
+        :param host: The Gradient host. By default it uses https://api.gradient.ai/.
+        """
+        gradientai_import.check()
+        self._host = host
+        self._model_name = model_name
+
+        self._gradient = Gradient(access_token=access_token, host=host, workspace_id=workspace_id)
+
+    def _get_telemetry_data(self) -> Dict[str, Any]:
+        """
+        Data that is sent to Posthog for usage analytics.
+        """
+        return {"model": self._model_name}
+
+    def to_dict(self) -> dict:
+        """
+        Serialize the component to a Python dictionary.
+        """
+        return default_to_dict(self, workspace_id=self._gradient.workspace_id, model_name=self._model_name)
+
+    def warm_up(self) -> None:
+        """
+        Load the embedding model.
+        """
+        if not hasattr(self, "_embedding_model"):
+            self._embedding_model = self._gradient.get_embeddings_model(slug=self._model_name)
+
+    @component.output_types(embedding=List[float])
+    def run(self, text: str):
+        """Generates an embedding for a single text."""
+        if not isinstance(text, str):
+            raise TypeError(
+                "GradientTextEmbedder expects a string as an input."
+                "In case you want to embed a list of Documents, please use the GradientDocumentEmbedder."
+            )
+
+        if not hasattr(self, "_embedding_model"):
+            raise RuntimeError("The embedding model has not been loaded. Please call warm_up() before running.")
+
+        result = self._embedding_model.generate_embeddings(inputs=[{"input": text}])
+
+        if (not result) or (result.embeddings is None) or (len(result.embeddings) == 0):
+            raise RuntimeError("The embedding model did not return any embeddings.")
+
+        return {"embedding": result.embeddings[0].embedding}

haystack/preview/components/generators/__init__.py

@@ -1,5 +1,6 @@
+from haystack.preview.components.generators.gradient.base import GradientGenerator
 from haystack.preview.components.generators.hugging_face_local import HuggingFaceLocalGenerator
 from haystack.preview.components.generators.hugging_face_tgi import HuggingFaceTGIGenerator
 from haystack.preview.components.generators.openai import GPTGenerator
 
-__all__ = ["HuggingFaceLocalGenerator", "HuggingFaceTGIGenerator", "GPTGenerator"]
+__all__ = ["GPTGenerator", "GradientGenerator", "HuggingFaceTGIGenerator", "HuggingFaceLocalGenerator"]