Retrieval Augmented Generation using Azure Databricks

Retrieval Augmented Generation (RAG) is a cutting-edge approach in AI that enhances large language models by integrating external knowledge sources. Azure Databricks offers a robust platform for developing RAG applications, allowing for the transformation of unstructured data into a format suitable for retrieval and response generation. This process involves a series of steps including understanding the user’s query, retrieving relevant data, and generating a response using a language model. The framework provided by Azure Databricks supports rapid iteration and deployment of RAG applications, ensuring high-quality, domain-specific responses that can include up-to-date information and proprietary knowledge.

This lab will take approximately 40 minutes to complete.

Before you start

You’ll need an Azure subscription in which you have administrative-level access.

Provision an Azure Databricks workspace

Tip: If you already have an Azure Databricks workspace, you can skip this procedure and use your existing workspace.

This exercise includes a script to provision a new Azure Databricks workspace. The script attempts to create a Premium tier Azure Databricks workspace resource in a region in which your Azure subscription has sufficient quota for the compute cores required in this exercise; and assumes your user account has sufficient permissions in the subscription to create an Azure Databricks workspace resource. If the script fails due to insufficient quota or permissions, you can try to create an Azure Databricks workspace interactively in the Azure portal.

  1. In a web browser, sign into the Azure portal at https://portal.azure.com.

  2. Use the [>_] button to the right of the search bar at the top of the page to create a new Cloud Shell in the Azure portal, selecting a PowerShell environment and creating storage if prompted. The cloud shell provides a command line interface in a pane at the bottom of the Azure portal, as shown here:

    Azure portal with a cloud shell pane

    Note: If you have previously created a cloud shell that uses a Bash environment, use the the drop-down menu at the top left of the cloud shell pane to change it to PowerShell.

  3. Note that you can resize the cloud shell by dragging the separator bar at the top of the pane, or by using the —, ◻, and X icons at the top right of the pane to minimize, maximize, and close the pane. For more information about using the Azure Cloud Shell, see the Azure Cloud Shell documentation.

  4. In the PowerShell pane, enter the following commands to clone this repo:

     rm -r mslearn-databricks -f
 git clone https://github.com/MicrosoftLearning/mslearn-databricks

  5. After the repo has been cloned, enter the following command to run the setup.ps1 script, which provisions an Azure Databricks workspace in an available region:

     ./mslearn-databricks/setup.ps1

  6. If prompted, choose which subscription you want to use (this will only happen if you have access to multiple Azure subscriptions).

  7. Wait for the script to complete - this typically takes around 5 minutes, but in some cases may take longer. While you are waiting, review the Introduction to Delta Lake article in the Azure Databricks documentation.

Create a cluster

Azure Databricks is a distributed processing platform that uses Apache Spark clusters to process data in parallel on multiple nodes. Each cluster consists of a driver node to coordinate the work, and worker nodes to perform processing tasks. In this exercise, you’ll create a single-node cluster to minimize the compute resources used in the lab environment (in which resources may be constrained). In a production environment, you’d typically create a cluster with multiple worker nodes.

Tip: If you already have a cluster with a 13.3 LTS ML or higher runtime version in your Azure Databricks workspace, you can use it to complete this exercise and skip this procedure.

  1. In the Azure portal, browse to the msl-xxxxxxx resource group that was created by the script (or the resource group containing your existing Azure Databricks workspace)
  2. Select your Azure Databricks Service resource (named databricks-xxxxxxx if you used the setup script to create it).

  3. In the Overview page for your workspace, use the Launch Workspace button to open your Azure Databricks workspace in a new browser tab; signing in if prompted.

    Tip: As you use the Databricks Workspace portal, various tips and notifications may be displayed. Dismiss these and follow the instructions provided to complete the tasks in this exercise.

  4. In the sidebar on the left, select the (+) New task, and then select Cluster.
  5. In the New Cluster page, create a new cluster with the following settings:
    • Cluster name: User Name’s cluster (the default cluster name)
    • Policy: Unrestricted
    • Cluster mode: Single Node
    • Access mode: Single user (with your user account selected)
    • Databricks runtime version: Select the ML edition of the latest non-beta version of the runtime (Not a Standard runtime version) that:
      • Does not use a GPU
      • Includes Scala > 2.11
      • Includes Spark > 3.4
    • Use Photon Acceleration: Unselected
    • Node type: Standard_D4ds_v5
    • Terminate after 20 minutes of inactivity
  6. Wait for the cluster to be created. It may take a minute or two.

Note: If your cluster fails to start, your subscription may have insufficient quota in the region where your Azure Databricks workspace is provisioned. See CPU core limit prevents cluster creation for details. If this happens, you can try deleting your workspace and creating a new one in a different region. You can specify a region as a parameter for the setup script like this: ./mslearn-databricks/setup.ps1 eastus

Install required libraries

  1. In your cluster’s page, select the Libraries tab.

  2. Select Install New.

  3. Select PyPI as the library source and type transformers==4.44.0 in the Package field.

  4. Select Install.

  5. Repeat the steps above to install databricks-vectorsearch==0.40as well.

Create a notebook and ingest data

  1. In the sidebar, use the (+) New link to create a Notebook. In the Connect drop-down list, select your cluster if it is not already selected. If the cluster is not running, it may take a minute or so to start.

  2. In the first cell of the notebook, enter the following code, which uses shell commands to download data files from GitHub into the file system used by your cluster.

     %sh
 rm -r /dbfs/RAG_lab
 mkdir /dbfs/RAG_lab
 wget -O /dbfs/RAG_lab/enwiki-latest-pages-articles.xml https://github.com/MicrosoftLearning/mslearn-databricks/raw/main/data/enwiki-latest-pages-articles.xml

  3. Use the ▸ Run Cell menu option at the left of the cell to run it. Then wait for the Spark job run by the code to complete.

  4. In a new cell, run the following code to create a dataframe from the raw data:

     from pyspark.sql import SparkSession

 # Create a Spark session
 spark = SparkSession.builder \
     .appName("RAG-DataPrep") \
     .getOrCreate()

 # Read the XML file
 raw_df = spark.read.format("xml") \
     .option("rowTag", "page") \
     .load("/RAG_lab/enwiki-latest-pages-articles.xml")

 # Show the DataFrame
 raw_df.show(5)

 # Print the schema of the DataFrame
 raw_df.printSchema()

  5. In a new cell, run the following code, replacing <catalog_name> with your Unity catalog’s name (the catalog with your workspace’s name plus a unique suffix), to clean and preprocess the data to extract the relevant text fields:

     from pyspark.sql.functions import col

 clean_df = raw_df.select(col("title"), col("revision.text._VALUE").alias("text"))
 clean_df = clean_df.na.drop()
 clean_df.write.format("delta").mode("overwrite").saveAsTable("<catalog_name>.default.wiki_pages")
 clean_df.show(5)

If you open the Catalog (CTRL + Alt + C) explorer and refresh its pane, you will see the Delta table created in your default Unity catalog.

Databricks’ Mosaic AI Vector Search is a vector database solution integrated within the Azure Databricks Platform. It optimizes the storage and retrieval of embeddings utilizing the Hierarchical Navigable Small World (HNSW) algorithm. It allows for efficient nearest neighbor searches, and its hybrid keyword-similarity search capability provides more relevant results by combining vector-based and keyword-based search techniques.

  1. In a new cell, run the following SQL query to enable the Change Data Feed feature in the source table before creating a delta sync index.

     %sql
 ALTER TABLE <catalog_name>.default.wiki_pages SET TBLPROPERTIES (delta.enableChangeDataFeed = true)

  2. In a new cell, run the following code to create the vector search index.

     from databricks.vector_search.client import VectorSearchClient

 client = VectorSearchClient()

 client.create_endpoint(
     name="vector_search_endpoint",
     endpoint_type="STANDARD"
 )

 index = client.create_delta_sync_index(
   endpoint_name="vector_search_endpoint",
   source_table_name="<catalog_name>.default.wiki_pages",
   index_name="<catalog_name>.default.wiki_index",
   pipeline_type="TRIGGERED",
   primary_key="title",
   embedding_source_column="text",
   embedding_model_endpoint_name="databricks-gte-large-en"
  )

If you open the Catalog (CTRL + Alt + C) explorer and refresh the its pane, you will see the index created in your default Unity catalog.

Note: Before running the next code cell, verify that the index was successfully created. To do that, right-click the index in the Catalog pane and select Open in Catalog Explorer. Wait until the index status is Online.

  1. In a new cell, run the following code to search for relevant documents based on a query vector.

     results_dict=index.similarity_search(
     query_text="Anthropology fields",
     columns=["title", "text"],
     num_results=1
 )

 display(results_dict)

Verify that the output finds the corresponding Wiki page related to the query prompt.

Augment Prompts with Retrieved Data:

Now we can enchance the capabilities of large language models by providing them with additional context from external data sources. By doing so, the models can generate more accurate and contextually relevant responses.

  1. In a new cell, run the following code to combine the retrieved data with the user’s query to create a rich prompt for the LLM.

     # Convert the dictionary to a DataFrame
 results = spark.createDataFrame([results_dict['result']['data_array'][0]])

 from transformers import pipeline

 # Load the summarization model
 summarizer = pipeline("summarization")

 # Extract the string values from the DataFrame column
 text_data = results.select("_2").rdd.flatMap(lambda x: x).collect()

 # Pass the extracted text data to the summarizer function
 summary = summarizer(text_data, max_length=512, min_length=100, do_sample=True)

 def augment_prompt(query_text):
     context = " ".join([item['summary_text'] for item in summary])
     return f"Query: {query_text}\nContext: {context}"

 prompt = augment_prompt("Explain the significance of Anthropology")
 print(prompt)

  2. In a new cell, run the following code to use an LLM to generate responses.

     from transformers import GPT2LMHeadModel, GPT2Tokenizer

 tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
 model = GPT2LMHeadModel.from_pretrained("gpt2")

 inputs = tokenizer(prompt, return_tensors="pt")
 outputs = model.generate(
     inputs["input_ids"], 
     max_length=300, 
     num_return_sequences=1, 
     repetition_penalty=2.0, 
     top_k=50, 
     top_p=0.95, 
     temperature=0.7,
     do_sample=True
 )
 response = tokenizer.decode(outputs[0], skip_special_tokens=True)

 print(response)

Clean up

In Azure Databricks portal, on the Compute page, select your cluster and select ■ Terminate to shut it down.

If you’ve finished exploring Azure Databricks, you can delete the resources you’ve created to avoid unnecessary Azure costs and free up capacity in your subscription.