Update text embedding component (#532)

PR that modifies the text embedding component: 
* Change the name of the component to `embed_text` for consistency (we
already have an `embed_image` component)
* Added VertexAI as a possible model to use 
* Changed the base image to pytorch since some of the dependencies in
the requirement have cuda deps
* Fixed some tests (mainly path related)

Tested the component with the CC rag pipeline and works fine (added the
[`op`](fcf8664)
here).

The component runs fine (exit code 0) however there seems to be an error
that pops since the connection with the API does not seem to shutdown
gracefully.

```
rag-cc-pipeline-embed_text-1     | [2023-10-18 13:58:45,227 | root | INFO] Writing data...
[########################################] | 100% Completed | 52.61 ss
rag-cc-pipeline-embed_text-1     | [2023-10-18 13:59:37,967 | fondant.executor | INFO] Saving output manifest to gs://soy-audio-379412_kfp-artifacts/custom_artifact/rag-cc-pipeline/rag-cc-pipeline-20231018155832/embed_text/manifest.json
rag-cc-pipeline-embed_text-1     | [2023-10-18 13:59:37,967 | fondant.executor | INFO] Writing cache key to gs://soy-audio-379412_kfp-artifacts/custom_artifact/rag-cc-pipeline/cache/dd3a24cb288cd0eaba12063d2885bc9d.txt
rag-cc-pipeline-embed_text-1     | Traceback (most recent call last):
rag-cc-pipeline-embed_text-1     |   File "src/python/grpcio/grpc/_cython/_cygrpc/aio/grpc_aio.pyx.pxi", line 110, in grpc._cython.cygrpc.shutdown_grpc_aio
rag-cc-pipeline-embed_text-1     |   File "src/python/grpcio/grpc/_cython/_cygrpc/aio/grpc_aio.pyx.pxi", line 114, in grpc._cython.cygrpc.shutdown_grpc_aio
rag-cc-pipeline-embed_text-1     |   File "src/python/grpcio/grpc/_cython/_cygrpc/aio/grpc_aio.pyx.pxi", line 78, in grpc._cython.cygrpc._actual_aio_shutdown
rag-cc-pipeline-embed_text-1     | AttributeError: 'NoneType' object has no attribute 'POLLER'
rag-cc-pipeline-embed_text-1     | Exception ignored in: 'grpc._cython.cygrpc.AioChannel.__dealloc__'
rag-cc-pipeline-embed_text-1     | Traceback (most recent call last):
rag-cc-pipeline-embed_text-1     |   File "src/python/grpcio/grpc/_cython/_cygrpc/aio/grpc_aio.pyx.pxi", line 110, in grpc._cython.cygrpc.shutdown_grpc_aio
rag-cc-pipeline-embed_text-1     |   File "src/python/grpcio/grpc/_cython/_cygrpc/aio/grpc_aio.pyx.pxi", line 114, in grpc._cython.cygrpc.shutdown_grpc_aio
rag-cc-pipeline-embed_text-1     |   File "src/python/grpcio/grpc/_cython/_cygrpc/aio/grpc_aio.pyx.pxi", line 78, in grpc._cython.cygrpc._actual_aio_shutdown
rag-cc-pipeline-embed_text-1     | AttributeError: 'NoneType' object has no attribute 'POLLER'
```

Normally this would be done with a client but in this case we're
initializing the model directly

components/generate_embeddings/Dockerfile → components/embed_text/Dockerfile

@@ -1,22 +1,30 @@
-FROM --platform=linux/amd64 python:3.8-slim as base
+FROM --platform=linux/amd64 pytorch/pytorch:2.0.1-cuda11.7-cudnn8-runtime as base
 
 # System dependencies
 RUN apt-get update && \
     apt-get upgrade -y && \
     apt-get install git -y
 
 # Install requirements
-COPY requirements.txt /
+COPY requirements.txt ./
 RUN pip3 install --no-cache-dir -r requirements.txt
 
 # Install Fondant
 # This is split from other requirements to leverage caching
 ARG FONDANT_VERSION=main
 RUN pip3 install fondant[aws,azure,gcp]@git+https://github.com/ml6team/fondant@${FONDANT_VERSION}
+
 # Set the working directory to the component folder
-WORKDIR /component/src
+WORKDIR /component
+COPY src/ src/
+ENV PYTHONPATH "${PYTHONPATH}:./src"
 
-# Copy over src-files
-COPY src/ .
+FROM base as test
+COPY test_requirements.txt .
+RUN pip3 install --no-cache-dir -r test_requirements.txt
+COPY tests/ tests/
+RUN python -m pytest tests
 
+FROM base
+WORKDIR /component/src
 ENTRYPOINT ["fondant", "execute", "main"]

components/generate_embeddings/README.md → components/embed_text/README.md

@@ -1,4 +1,4 @@
-# Generate embeddings
+# Embed text
 
 ### Description
 Component that generates embeddings of text passages.
@@ -22,9 +22,10 @@ The component takes the following arguments to alter its behavior:
 
 | argument | type | description | default |
 | -------- | ---- | ----------- | ------- |
-| model_provider | str | The provider of the model - corresponding to langchain embedding classes. Currently the following providers are supported: aleph_alpha, cohere, huggingface, openai. | huggingface |
-| model | str | The model to generate embeddings from. Choose an available model name to pass to the model provider's langchain embedding class. | all-MiniLM-L6-v2 |
+| model_provider | str | The provider of the model - corresponding to langchain embedding classes. Currently the following providers are supported: aleph_alpha, cohere, huggingface, openai, vertexai. | huggingface |
+| model | str | The model to generate embeddings from. Choose an available model name to pass to the model provider's langchain embedding class. | / |
 | api_keys | dict | The API keys to use for the model provider that are written to environment variables.Pass only the keys required by the model provider or conveniently pass all keys you will ever need. Pay attention how to name the dictionary keys so that they can be used by the model provider. | / |
+| auth_kwargs | dict | Additional keyword arguments required for api initialization/authentication. | / |
 
 ### Usage
 
@@ -34,15 +35,21 @@ You can add this component to your pipeline using the following code:
 from fondant.pipeline import ComponentOp
 
 
-generate_embeddings_op = ComponentOp.from_registry(
-    name="generate_embeddings",
+embed_text_op = ComponentOp.from_registry(
+    name="embed_text",
     arguments={
-        # Add arguments
         # "model_provider": "huggingface",
-        # "model": "all-MiniLM-L6-v2",
+        # "model": ,
         # "api_keys": {},
+        # "auth_kwargs": {},
     }
 )
-pipeline.add_op(generate_embeddings_op, dependencies=[...])  #Add previous component as dependency
+pipeline.add_op(embed_text_op, dependencies=[...])  #Add previous component as dependency
 ```
 
+### Testing
+
+You can run the tests using docker with BuildKit. From this directory, run:
+```
+docker build . --target test
+```

components/generate_embeddings/fondant_component.yaml → components/embed_text/fondant_component.yaml

@@ -1,6 +1,6 @@
-name: Generate embeddings
+name: Embed text
 description: Component that generates embeddings of text passages.
-image: generate_embeddings:latest
+image: embed_text:latest
 
 consumes:
   text:
@@ -22,20 +22,28 @@ args:
   model_provider:
     description: |
       The provider of the model - corresponding to langchain embedding classes. 
-      Currently the following providers are supported: aleph_alpha, cohere, huggingface, openai.
+      Currently the following providers are supported: aleph_alpha, cohere, huggingface, openai, 
+      vertexai.
     type: str
     default: huggingface
   model:
     description: |
       The model to generate embeddings from. 
       Choose an available model name to pass to the model provider's langchain embedding class.
     type: str
-    default: all-MiniLM-L6-v2
+    default: None
   api_keys:
     description: |
       The API keys to use for the model provider that are written to environment variables.
       Pass only the keys required by the model provider or conveniently pass all keys you will ever need. 
       Pay attention how to name the dictionary keys so that they can be used by the model provider.
     type: dict
+    default: {}
+  auth_kwargs:
+    description: |
+      Additional keyword arguments required for api initialization/authentication.
+    type: dict
+    default: {}
+
 
 

components/generate_embeddings/requirements.txt → components/embed_text/requirements.txt

@@ -1,5 +1,6 @@
 aleph_alpha_client==3.5.1
 cohere==4.27
+google-cloud-aiplatform==1.34.0
 langchain==0.0.313
 openai==0.28.1
 pandas==1.5.0

components/generate_embeddings/src/main.py → components/embed_text/src/main.py

@@ -1,33 +1,53 @@
 import logging
+import os
 
+import google.cloud.aiplatform as aip
 import pandas as pd
 from fondant.component import PandasTransformComponent
 from langchain.embeddings import (
     AlephAlphaAsymmetricSemanticEmbedding,
     CohereEmbeddings,
     HuggingFaceEmbeddings,
     OpenAIEmbeddings,
+    VertexAIEmbeddings,
 )
+from langchain.schema.embeddings import Embeddings
 from retry import retry
-from utils import to_env_vars
 
 logger = logging.getLogger(__name__)
 
 
-class GenerateEmbeddingsComponent(PandasTransformComponent):
+def to_env_vars(api_keys: dict):
+    for key, value in api_keys.items():
+        os.environ[key] = value
+
+
+class EmbedTextComponent(PandasTransformComponent):
     def __init__(
         self,
         *_,
         model_provider: str,
         model: str,
         api_keys: dict,
+        auth_kwargs: dict,
     ):
-        self.model_provider = model_provider
-        self.model = model
+        self.embedding_model = self.get_embedding_model(
+            model_provider,
+            model,
+            auth_kwargs,
+        )
 
         to_env_vars(api_keys)
 
-    def get_embedding_model(self, model_provider, model: str):
+    @staticmethod
+    def get_embedding_model(
+        model_provider,
+        model: str,
+        auth_kwargs: dict,
+    ) -> Embeddings:
+        if model_provider == "vertexai":
+            aip.init(**auth_kwargs)
+            return VertexAIEmbeddings(model=model)
         # contains a first selection of embedding models
         if model_provider == "aleph_alpha":
             return AlephAlphaAsymmetricSemanticEmbedding(model=model)
@@ -41,13 +61,11 @@ def get_embedding_model(self, model_provider, model: str):
         raise ValueError(msg)
 
     @retry()  # make sure to keep trying even when api call limit is reached
-    def get_embeddings_vectors(self, embedding_model, texts):
-        return embedding_model.embed_documents(texts.tolist())
+    def get_embeddings_vectors(self, texts):
+        return self.embedding_model.embed_documents(texts.tolist())
 
     def transform(self, dataframe: pd.DataFrame) -> pd.DataFrame:
-        embedding_model = self.get_embedding_model(self.model_provider, self.model)
         dataframe[("text", "embedding")] = self.get_embeddings_vectors(
-            embedding_model,
             dataframe[("text", "data")],
         )
         return dataframe

components/embed_text/test_requirements.txt

@@ -0,0 +1 @@
+pytest==7.4.2

components/generate_embeddings/tests/generate_embeddings_test.py → components/embed_text/tests/embed_text_test.py

@@ -4,7 +4,7 @@
 
 import pandas as pd
 
-from components.generate_embeddings.src.main import GenerateEmbeddingsComponent
+from src.main import EmbedTextComponent
 
 
 def embeddings_close(a, b):
@@ -13,9 +13,9 @@ def embeddings_close(a, b):
 
 def test_run_component_test():
     """Test generate embeddings component."""
-    with open("lorem_300.txt", encoding="utf-8") as f:
+    with open("tests/lorem_300.txt", encoding="utf-8") as f:
         lorem_300 = f.read()
-    with open("lorem_400.txt", encoding="utf-8") as f:
+    with open("tests/lorem_400.txt", encoding="utf-8") as f:
         lorem_400 = f.read()
 
     # Given: Dataframe with text
@@ -29,15 +29,16 @@ def test_run_component_test():
 
     dataframe = pd.concat({"text": pd.DataFrame(data)}, axis=1, names=["text", "data"])
 
-    component = GenerateEmbeddingsComponent(
+    component = EmbedTextComponent(
         model_provider="huggingface",
         model="all-MiniLM-L6-v2",
         api_keys={},
+        auth_kwargs={},
     )
 
     dataframe = component.transform(dataframe=dataframe)
 
-    with open("hello_world_embedding.txt", encoding="utf-8") as f:
+    with open("tests/hello_world_embedding.txt", encoding="utf-8") as f:
         hello_world_embedding = f.read()
         hello_world_embedding = json.loads(hello_world_embedding)