Merge pull request #104 from Azure-Samples/test-eval3

Add comment in workflow

Author: pamelafox

.github/workflows/evaluate.yaml

@@ -46,6 +46,17 @@ jobs:
         run: |
           echo "Comment contains #evaluate hashtag"
 
+      - name: Comment on pull request
+        uses: actions/github-script@v7
+        with:
+          script: |
+            github.rest.issues.createComment({
+              issue_number: context.issue.number,
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              body: "Starting evaluation! Check the Actions tab for progress, or wait for a comment with the results."
+            })
+
       - uses: actions/checkout@v4
 
       - name: Install pgvector
@@ -133,7 +144,7 @@ jobs:
 
       - name: Evaluate local RAG flow
         run: |
-          python evals/evaluate.py --targeturl=http://127.0.0.1:8000/chat --numquestions=2 --resultsdir=evals/results/pr${{ github.event.issue.number }}
+          python evals/evaluate.py --targeturl=http://127.0.0.1:8000/chat --resultsdir=evals/results/pr${{ github.event.issue.number }}
 
       - name: Upload server logs as build artifact
         uses: actions/upload-artifact@v4

docs/evaluation.md

@@ -33,17 +33,19 @@ pip install -r requirements-dev.txt
 
 ## Generate ground truth data
 
+Modify the prompt in `evals/generate.txt` to match your database table and RAG scenario.
+
 Generate ground truth data by running the following command:
 
 ```bash
-python evals/generate.py
+python evals/generate_ground_truth_data.py
 ```
 
 Review the generated data after running that script, removing any question/answer pairs that don't seem like realistic user input.
 
 ## Evaluate the RAG answer quality
 
-Review the configuration in `evals/eval_config.json` to ensure that everything is correctly setup. You may want to adjust the metrics used. [TODO: link to evaluator docs]
+Review the configuration in `evals/eval_config.json` to ensure that everything is correctly setup. You may want to adjust the metrics used. See [the ai-rag-chat-evaluator README](https://github.com/Azure-Samples/ai-rag-chat-evaluator) for more information on the available metrics.
 
 By default, the evaluation script will evaluate every question in the ground truth data.
 Run the evaluation script by running the following command:
@@ -68,8 +70,6 @@ Compare answers across runs by running the following command:
 python -m evaltools diff evals/results/baseline/
 ```
 
-## Run the evaluation in GitHub actions
-
+## Run the evaluation on a PR
 
-# TODO: Add GPT-4 deployment with high capacity for evaluation
-# TODO: Add CI workflow that can be triggered to run the evaluate on the local app
+To run the evaluation on the changes in a PR, you can add a `/evaluate` comment to the PR. This will trigger the evaluation workflow to run the evaluation on the PR changes, and will post the results to the PR.

evals/generate_ground_truth.py

@@ -1,11 +1,13 @@
+import json
 import logging
 import os
 from collections.abc import Generator
 from pathlib import Path
 
-from azure.identity import AzureDeveloperCliCredential
-from dotenv import load_dotenv
-from evaltools.gen.generate import generate_test_qa_data
+from azure.identity import AzureDeveloperCliCredential, get_bearer_token_provider
+from dotenv_azd import load_azd_env
+from openai import AzureOpenAI, OpenAI
+from openai.types.chat import ChatCompletionToolParam
 from sqlalchemy import create_engine, select
 from sqlalchemy.orm import Session
 
@@ -14,7 +16,37 @@
 logger = logging.getLogger("ragapp")
 
 
-def source_retriever() -> Generator[dict, None, None]:
+def qa_pairs_tool(num_questions: int = 1) -> ChatCompletionToolParam:
+    return {
+        "type": "function",
+        "function": {
+            "name": "qa_pairs",
+            "description": "Send in question and answer pairs for a customer-facing chat app",
+            "parameters": {
+                "type": "object",
+                "properties": {
+                    "qa_list": {
+                        "type": "array",
+                        "description": f"List of {num_questions} question and answer pairs",
+                        "items": {
+                            "type": "object",
+                            "properties": {
+                                "question": {"type": "string", "description": "The question text"},
+                                "answer": {"type": "string", "description": "The answer text"},
+                            },
+                            "required": ["question", "answer"],
+                        },
+                        "minItems": num_questions,
+                        "maxItems": num_questions,
+                    }
+                },
+                "required": ["qa_list"],
+            },
+        },
+    }
+
+
+def source_retriever() -> Generator[str, None, None]:
     # Connect to the database
     DBHOST = os.environ["POSTGRES_HOST"]
     DBUSER = os.environ["POSTGRES_USERNAME"]
@@ -27,16 +59,14 @@ def source_retriever() -> Generator[dict, None, None]:
         item_types = session.scalars(select(Item.type).distinct())
         for item_type in item_types:
             records = list(session.scalars(select(Item).filter(Item.type == item_type).order_by(Item.id)))
-            # logger.info(f"Processing database records for type: {item_type}")
-            # yield {
-            #    "citations": " ".join([f"[{record.id}] - {record.name}" for record in records]),
-            #    "content": "\n\n".join([record.to_str_for_rag() for record in records]),
-            # }
+            logger.info(f"Processing database records for type: {item_type}")
+            yield "\n\n".join([f"## Product ID: [{record.id}]\n" + record.to_str_for_rag() for record in records])
         # Fetch each item individually
-        records = list(session.scalars(select(Item).order_by(Item.id)))
-        for record in records:
-            logger.info(f"Processing database record: {record.name}")
-            yield {"id": record.id, "content": record.to_str_for_rag()}
+        # records = list(session.scalars(select(Item).order_by(Item.id)))
+        # for record in records:
+        #    logger.info(f"Processing database record: {record.name}")
+        #    yield f"## Product ID: [{record.id}]\n" + record.to_str_for_rag()
+        # await self.openai_chat_client.chat.completions.create(
 
 
 def source_to_text(source) -> str:
@@ -47,49 +77,76 @@ def answer_formatter(answer, source) -> str:
     return f"{answer} [{source['id']}]"
 
 
-def get_openai_config_dict() -> dict:
-    """Return a dictionary with OpenAI configuration based on environment variables."""
+def get_openai_client() -> tuple[AzureOpenAI | OpenAI, str]:
+    """Return an OpenAI client based on the environment variables"""
+    openai_client: AzureOpenAI | OpenAI
     OPENAI_CHAT_HOST = os.getenv("OPENAI_CHAT_HOST")
     if OPENAI_CHAT_HOST == "azure":
         if api_key := os.getenv("AZURE_OPENAI_KEY"):
             logger.info("Using Azure OpenAI Service with API Key from AZURE_OPENAI_KEY")
-            api_key = os.environ["AZURE_OPENAI_KEY"]
+            openai_client = AzureOpenAI(
+                api_version=os.environ["AZURE_OPENAI_VERSION"],
+                azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
+                api_key=api_key,
+            )
         else:
             logger.info("Using Azure OpenAI Service with Azure Developer CLI Credential")
-            azure_credential = AzureDeveloperCliCredential(process_timeout=60)
-            api_key = azure_credential.get_token("https://cognitiveservices.azure.com/.default").token
-        openai_config = {
-            "api_type": "azure",
-            "api_base": os.environ["AZURE_OPENAI_ENDPOINT"],
-            "api_key": api_key,
-            "api_version": os.environ["AZURE_OPENAI_VERSION"],
-            "deployment": os.environ["AZURE_OPENAI_CHAT_DEPLOYMENT"],
-            "model": os.environ["AZURE_OPENAI_CHAT_DEPLOYMENT"],
-        }
+            azure_credential = AzureDeveloperCliCredential(process_timeout=60, tenant_id=os.environ["AZURE_TENANT_ID"])
+            token_provider = get_bearer_token_provider(azure_credential, "https://cognitiveservices.azure.com/.default")
+            openai_client = AzureOpenAI(
+                api_version=os.environ["AZURE_OPENAI_VERSION"],
+                azure_endpoint=os.environ["AZURE_OPENAI_ENDPOINT"],
+                azure_ad_token_provider=token_provider,
+            )
+        model = os.environ["AZURE_OPENAI_CHAT_DEPLOYMENT"]
     elif OPENAI_CHAT_HOST == "ollama":
         raise NotImplementedError("Ollama OpenAI Service is not supported. Switch to Azure or OpenAI.com")
     else:
         logger.info("Using OpenAI Service with API Key from OPENAICOM_KEY")
-        openai_config = {
-            "api_type": "openai",
-            "api_key": os.environ["OPENAICOM_KEY"],
-            "model": os.environ["OPENAICOM_CHAT_MODEL"],
-            "deployment": "none-needed-for-openaicom",
-        }
-    return openai_config
+        openai_client = OpenAI(api_key=os.environ["OPENAICOM_KEY"])
+        model = os.environ["OPENAICOM_CHAT_MODEL"]
+    return openai_client, model
+
+
+def generate_ground_truth_data(num_questions_total: int, num_questions_per_source: int = 5):
+    logger.info("Generating %d questions total", num_questions_total)
+    openai_client, model = get_openai_client()
+    current_dir = Path(__file__).parent
+    generate_prompt = open(current_dir / "generate_prompt.txt").read()
+    output_file = Path(__file__).parent / "ground_truth.jsonl"
+
+    qa: list[dict] = []
+    for source in source_retriever():
+        if len(qa) > num_questions_total:
+            logger.info("Generated enough questions already, stopping")
+            break
+        result = openai_client.chat.completions.create(
+            model=model,
+            messages=[
+                {"role": "system", "content": generate_prompt},
+                {"role": "user", "content": json.dumps(source)},
+            ],
+            tools=[qa_pairs_tool(num_questions=2)],
+        )
+        if not result.choices[0].message.tool_calls:
+            logger.warning("No tool calls found in response, skipping")
+            continue
+        qa_pairs = json.loads(result.choices[0].message.tool_calls[0].function.arguments)["qa_list"]
+        qa_pairs = [{"question": qa_pair["question"], "truth": qa_pair["answer"]} for qa_pair in qa_pairs]
+        qa.extend(qa_pairs)
+
+    logger.info("Writing %d questions to %s", num_questions_total, output_file)
+    directory = Path(output_file).parent
+    if not directory.exists():
+        directory.mkdir(parents=True)
+    with open(output_file, "w", encoding="utf-8") as f:
+        for item in qa[0:num_questions_total]:
+            f.write(json.dumps(item) + "\n")
 
 
 if __name__ == "__main__":
     logging.basicConfig(level=logging.WARNING)
     logger.setLevel(logging.INFO)
-    load_dotenv(".env", override=True)
-
-    generate_test_qa_data(
-        openai_config=get_openai_config_dict(),
-        num_questions_total=202,
-        num_questions_per_source=2,
-        output_file=Path(__file__).parent / "ground_truth.jsonl",
-        source_retriever=source_retriever,
-        source_to_text=source_to_text,
-        answer_formatter=answer_formatter,
-    )
+    load_azd_env()
+
+    generate_ground_truth_data(num_questions_total=10)

evals/generate_prompt.txt

@@ -0,0 +1,9 @@
+Your job is to generate example questions that a customer might ask about the products.
+You should come up with a question and an answer based on the provided data.
+The answer should include the product ID in square brackets.
+For example,
+'What climbing gear do you have?'
+with answer:
+'We have a variety of climbing gear, including ropes, harnesses, and carabiners. [1][2]'.
+Remember that customers probably don't know the names of specific brands,
+so your questions should be more general questions from someone who is shopping for these types of products.