fix: Psl 6635 - BYOD workflow implementation (#1265)

Co-authored-by: Ajit Padhi (Persistent Systems Inc) <v-padhiajit@microsoft.com>
Co-authored-by: Francia Riesco <Fr4nc3@users.noreply.github.com>

Author: 3 people

code/backend/Admin.py

@@ -51,6 +51,7 @@ def load_css(file_path):
     """
          * If you want to ingest data (pdf, websites, etc.), then use the `Ingest Data` tab
          * If you want to explore how your data was chunked, check the `Explore Data` tab
+         * If you want to delete your data, check the `Delete Data` tab
          * If you want to adapt the underlying prompts, logging settings and others, use the `Configuration` tab
          """
 )

code/backend/batch/utilities/helpers/config/config_helper.py

@@ -12,6 +12,7 @@
 from ...orchestrator import OrchestrationSettings
 from ..env_helper import EnvHelper
 from .assistant_strategy import AssistantStrategy
+from .conversation_flow import ConversationFlow
 
 CONFIG_CONTAINER_NAME = "config"
 CONFIG_FILE_NAME = "active.json"
@@ -90,6 +91,9 @@ def get_available_orchestration_strategies(self):
     def get_available_ai_assistant_types(self):
         return [c.value for c in AssistantStrategy]
 
+    def get_available_conversational_flows(self):
+        return [c.value for c in ConversationFlow]
+
 
 # TODO: Change to AnsweringChain or something, Prompts is not a good name
 class Prompts:
@@ -102,6 +106,7 @@ def __init__(self, prompts: dict):
         self.enable_post_answering_prompt = prompts["enable_post_answering_prompt"]
         self.enable_content_safety = prompts["enable_content_safety"]
         self.ai_assistant_type = prompts["ai_assistant_type"]
+        self.conversational_flow = prompts["conversational_flow"]
 
 
 class Example:
@@ -166,13 +171,20 @@ def _set_new_config_properties(config: dict, default_config: dict):
             config["example"] = default_config["example"]
 
         if config["prompts"].get("ai_assistant_type") is None:
-            config["prompts"]["ai_assistant_type"] = default_config["prompts"]["ai_assistant_type"]
+            config["prompts"]["ai_assistant_type"] = default_config["prompts"][
+                "ai_assistant_type"
+            ]
 
         if config.get("integrated_vectorization_config") is None:
             config["integrated_vectorization_config"] = default_config[
                 "integrated_vectorization_config"
             ]
 
+        if config["prompts"].get("conversational_flow") is None:
+            config["prompts"]["conversational_flow"] = default_config["prompts"][
+                "conversational_flow"
+            ]
+
     @staticmethod
     @functools.cache
     def get_active_config_or_default():
@@ -247,12 +259,14 @@ def get_default_config():
     @staticmethod
     @functools.cache
     def get_default_contract_assistant():
-        contract_file_path = os.path.join(os.path.dirname(__file__), "default_contract_assistant_prompt.txt")
+        contract_file_path = os.path.join(
+            os.path.dirname(__file__), "default_contract_assistant_prompt.txt"
+        )
         contract_assistant = ""
         with open(contract_file_path, encoding="utf-8") as f:
             contract_assistant = f.readlines()
 
-        return ''.join([str(elem) for elem in contract_assistant])
+        return "".join([str(elem) for elem in contract_assistant])
 
     @staticmethod
     def clear_config():

code/backend/batch/utilities/helpers/config/default.json

@@ -8,7 +8,8 @@
     "use_on_your_data_format": true,
     "enable_post_answering_prompt": false,
     "ai_assistant_type": "default",
-    "enable_content_safety": true
+    "enable_content_safety": true,
+    "conversational_flow": "custom"
   },
   "example": {
     "documents": "{\n  \"retrieved_documents\": [\n    {\n      \"[doc1]\": {\n        \"content\": \"Dual Transformer Encoder (DTE) DTE (https://dev.azure.com/TScience/TSciencePublic/_wiki/wikis/TSciencePublic.wiki/82/Dual-Transformer-Encoder) DTE is a general pair-oriented sentence representation learning framework based on transformers. It provides training, inference and evaluation for sentence similarity models. Model Details DTE can be used to train a model for sentence similarity with the following features: - Build upon existing transformer-based text representations (e.g.TNLR, BERT, RoBERTa, BAG-NLR) - Apply smoothness inducing technology to improve the representation robustness - SMART (https://arxiv.org/abs/1911.03437) SMART - Apply NCE (Noise Contrastive Estimation) based similarity learning to speed up training of 100M pairs We use pretrained DTE model\"\n      }\n    },\n    {\n      \"[doc2]\": {\n        \"content\": \"trained on internal data. You can find more details here - Models.md (https://dev.azure.com/TScience/_git/TSciencePublic?path=%2FDualTransformerEncoder%2FMODELS.md&version=GBmaster&_a=preview) Models.md DTE-pretrained for In-context Learning Research suggests that finetuned transformers can be used to retrieve semantically similar exemplars for e.g. KATE (https://arxiv.org/pdf/2101.06804.pdf) KATE . They show that finetuned models esp. tuned on related tasks give the maximum boost to GPT-3 in-context performance. DTE have lot of pretrained models that are trained on intent classification tasks. We can use these model embedding to find natural language utterances which are similar to our test utterances at test time. The steps are: 1. Embed\"\n      }\n    },\n    {\n      \"[doc3]\": {\n        \"content\": \"train and test utterances using DTE model 2. For each test embedding, find K-nearest neighbors. 3. Prefix the prompt with nearest embeddings. The following diagram from the above paper (https://arxiv.org/pdf/2101.06804.pdf) the above paper visualizes this process: DTE-Finetuned This is an extension of DTE-pretrained method where we further finetune the embedding models for prompt crafting task. In summary, we sample random prompts from our training data and use them for GPT-3 inference for the another part of training data. Some prompts work better and lead to right results whereas other prompts lead\"\n      }\n    },\n    {\n      \"[doc4]\": {\n        \"content\": \"to wrong completions. We finetune the model on the downstream task of whether a prompt is good or not based on whether it leads to right or wrong completion. This approach is similar to this paper: Learning To Retrieve Prompts for In-Context Learning (https://arxiv.org/pdf/2112.08633.pdf) this paper: Learning To Retrieve Prompts for In-Context Learning . This method is very general but it may require a lot of data to actually finetune a model to learn how to retrieve examples suitable for the downstream inference model like GPT-3.\"\n      }\n    }\n  ]\n}",

code/backend/batch/utilities/helpers/env_helper.py

@@ -4,7 +4,6 @@
 from dotenv import load_dotenv
 from azure.identity import DefaultAzureCredential, get_bearer_token_provider
 from azure.keyvault.secrets import SecretClient
-from .config.conversation_flow import ConversationFlow
 
 logger = logging.getLogger(__name__)
 
@@ -69,9 +68,13 @@ def __load_config(self, **kwargs) -> None:
         self.AZURE_SEARCH_FIELDS_METADATA = os.getenv(
             "AZURE_SEARCH_FIELDS_METADATA", "metadata"
         )
-        self.AZURE_SEARCH_SOURCE_COLUMN = os.getenv("AZURE_SEARCH_SOURCE_COLUMN", "source")
+        self.AZURE_SEARCH_SOURCE_COLUMN = os.getenv(
+            "AZURE_SEARCH_SOURCE_COLUMN", "source"
+        )
         self.AZURE_SEARCH_CHUNK_COLUMN = os.getenv("AZURE_SEARCH_CHUNK_COLUMN", "chunk")
-        self.AZURE_SEARCH_OFFSET_COLUMN = os.getenv("AZURE_SEARCH_OFFSET_COLUMN", "offset")
+        self.AZURE_SEARCH_OFFSET_COLUMN = os.getenv(
+            "AZURE_SEARCH_OFFSET_COLUMN", "offset"
+        )
         self.AZURE_SEARCH_CONVERSATIONS_LOG_INDEX = os.getenv(
             "AZURE_SEARCH_CONVERSATIONS_LOG_INDEX", "conversations"
         )
@@ -211,10 +214,6 @@ def __load_config(self, **kwargs) -> None:
         self.ORCHESTRATION_STRATEGY = os.getenv(
             "ORCHESTRATION_STRATEGY", "openai_function"
         )
-        # Conversation Type - which chooses between custom or byod
-        self.CONVERSATION_FLOW = os.getenv(
-            "CONVERSATION_FLOW", ConversationFlow.CUSTOM.value
-        )
         # Speech Service
         self.AZURE_SPEECH_SERVICE_NAME = os.getenv("AZURE_SPEECH_SERVICE_NAME", "")
         self.AZURE_SPEECH_SERVICE_REGION = os.getenv("AZURE_SPEECH_SERVICE_REGION")

code/backend/pages/04_Configuration.py

@@ -7,6 +7,7 @@
 from batch.utilities.helpers.config.config_helper import ConfigHelper
 from azure.core.exceptions import ResourceNotFoundError
 from batch.utilities.helpers.config.assistant_strategy import AssistantStrategy
+from batch.utilities.helpers.config.conversation_flow import ConversationFlow
 
 sys.path.append(os.path.join(os.path.dirname(__file__), ".."))
 env_helper: EnvHelper = EnvHelper()
@@ -65,6 +66,8 @@ def load_css(file_path):
     st.session_state["orchestrator_strategy"] = config.orchestrator.strategy.value
 if "ai_assistant_type" not in st.session_state:
     st.session_state["ai_assistant_type"] = config.prompts.ai_assistant_type
+if "conversational_flow" not in st.session_state:
+    st.session_state["conversational_flow"] = config.prompts.conversational_flow
 
 if env_helper.AZURE_SEARCH_USE_INTEGRATED_VECTORIZATION:
     if "max_page_length" not in st.session_state:
@@ -163,13 +166,30 @@ def validate_documents():
 
 
 try:
+    conversational_flow_help = "Whether to use the custom conversational flow or byod conversational flow. Refer to the Conversational flow options README for more details."
+    with st.expander("Conversational flow configuration", expanded=True):
+        cols = st.columns([2, 4])
+        with cols[0]:
+            conv_flow = st.selectbox(
+                "Conversational flow",
+                key="conversational_flow",
+                options=config.get_available_conversational_flows(),
+                help=conversational_flow_help,
+            )
+
     with st.expander("Orchestrator configuration", expanded=True):
         cols = st.columns([2, 4])
         with cols[0]:
             st.selectbox(
                 "Orchestrator strategy",
                 key="orchestrator_strategy",
                 options=config.get_available_orchestration_strategies(),
+                disabled=(
+                    True
+                    if st.session_state["conversational_flow"]
+                    == ConversationFlow.BYOD.value
+                    else False
+                ),
             )
 
     # # # condense_question_prompt_help = "This prompt is used to convert the user's input to a standalone question, using the context of the chat history."
@@ -377,6 +397,7 @@ def validate_documents():
                 ],
                 "enable_content_safety": st.session_state["enable_content_safety"],
                 "ai_assistant_type": st.session_state["ai_assistant_type"],
+                "conversational_flow": st.session_state["conversational_flow"],
             },
             "messages": {
                 "post_answering_filter": st.session_state[

code/create_app.py

@@ -8,23 +8,67 @@
 import mimetypes
 from os import path
 import sys
+import re
 import requests
 from openai import AzureOpenAI, Stream, APIStatusError
 from openai.types.chat import ChatCompletionChunk
 from flask import Flask, Response, request, Request, jsonify
 from dotenv import load_dotenv
+from urllib.parse import quote
 from backend.batch.utilities.helpers.env_helper import EnvHelper
 from backend.batch.utilities.helpers.orchestrator_helper import Orchestrator
 from backend.batch.utilities.helpers.config.config_helper import ConfigHelper
 from backend.batch.utilities.helpers.config.conversation_flow import ConversationFlow
 from azure.mgmt.cognitiveservices import CognitiveServicesManagementClient
 from azure.identity import DefaultAzureCredential
+from backend.batch.utilities.helpers.azure_blob_storage_client import (
+    AzureBlobStorageClient,
+)
 
 ERROR_429_MESSAGE = "We're currently experiencing a high number of requests for the service you're trying to access. Please wait a moment and try again."
 ERROR_GENERIC_MESSAGE = "An error occurred. Please try again. If the problem persists, please contact the site administrator."
 logger = logging.getLogger(__name__)
 
 
+def get_markdown_url(source, title, container_sas):
+    """Get Markdown URL for a citation"""
+
+    url = quote(source, safe=":/")
+    if "_SAS_TOKEN_PLACEHOLDER_" in url:
+        url = url.replace("_SAS_TOKEN_PLACEHOLDER_", container_sas)
+    return f"[{title}]({url})"
+
+
+def get_citations(citation_list):
+    """Returns Formated Citations"""
+    blob_client = AzureBlobStorageClient()
+    container_sas = blob_client.get_container_sas()
+    citations_dict = {"citations": []}
+    for citation in citation_list.get("citations"):
+        metadata = (
+            json.loads(citation["url"])
+            if isinstance(citation["url"], str)
+            else citation["url"]
+        )
+        title = citation["title"]
+        url = get_markdown_url(metadata["source"], title, container_sas)
+        citations_dict["citations"].append(
+            {
+                "content": url + "\n\n\n" + citation["content"],
+                "id": metadata["id"],
+                "chunk_id": (
+                    re.findall(r"\d+", metadata["chunk_id"])[-1]
+                    if metadata["chunk_id"] is not None
+                    else metadata["chunk"]
+                ),
+                "title": title,
+                "filepath": title.split("/")[-1],
+                "url": url,
+            }
+        )
+    return citations_dict
+
+
 def stream_with_data(response: Stream[ChatCompletionChunk]):
     """This function streams the response from Azure OpenAI with data."""
     response_obj = {
@@ -67,8 +111,9 @@ def stream_with_data(response: Stream[ChatCompletionChunk]):
         role = delta.role
 
         if role == "assistant":
+            citations = get_citations(delta.model_extra["context"])
             response_obj["choices"][0]["messages"][0]["content"] = json.dumps(
-                delta.model_extra["context"],
+                citations,
                 ensure_ascii=False,
             )
         else:
@@ -135,7 +180,8 @@ def conversation_with_data(conversation: Request, env_helper: EnvHelper):
                                 env_helper.AZURE_SEARCH_CONTENT_VECTOR_COLUMN
                             ],
                             "title_field": env_helper.AZURE_SEARCH_TITLE_COLUMN or None,
-                            "url_field": env_helper.AZURE_SEARCH_URL_COLUMN or None,
+                            "url_field": env_helper.AZURE_SEARCH_FIELDS_METADATA
+                            or None,
                             "filepath_field": (
                                 env_helper.AZURE_SEARCH_FILENAME_COLUMN or None
                             ),
@@ -166,6 +212,7 @@ def conversation_with_data(conversation: Request, env_helper: EnvHelper):
     )
 
     if not env_helper.SHOULD_STREAM:
+        citations = get_citations(response.choices[0].message.model_extra["context"])
         response_obj = {
             "id": response.id,
             "model": response.model,
@@ -176,7 +223,7 @@ def conversation_with_data(conversation: Request, env_helper: EnvHelper):
                     "messages": [
                         {
                             "content": json.dumps(
-                                response.choices[0].message.model_extra["context"],
+                                citations,
                                 ensure_ascii=False,
                             ),
                             "end_turn": False,
@@ -194,10 +241,7 @@ def conversation_with_data(conversation: Request, env_helper: EnvHelper):
 
         return response_obj
 
-    return Response(
-        stream_with_data(response),
-        mimetype="application/json-lines",
-    )
+    return Response(stream_with_data(response), mimetype="application/json-lines")
 
 
 def stream_without_data(response: Stream[ChatCompletionChunk]):
@@ -405,7 +449,9 @@ async def conversation_custom():
 
     @app.route("/api/conversation", methods=["POST"])
     async def conversation():
-        conversation_flow = env_helper.CONVERSATION_FLOW
+        ConfigHelper.get_active_config_or_default.cache_clear()
+        result = ConfigHelper.get_active_config_or_default()
+        conversation_flow = result.prompts.conversational_flow
         if conversation_flow == ConversationFlow.CUSTOM.value:
             return await conversation_custom()
         elif conversation_flow == ConversationFlow.BYOD.value:

code/frontend/src/components/Answer/AnswerParser.tsx

@@ -11,7 +11,7 @@ let filteredCitations = [] as Citation[];
 
 // Define a function to check if a citation with the same Chunk_Id already exists in filteredCitations
 const isDuplicate = (citation: Citation,citationIndex:string) => {
-    return filteredCitations.some((c) => c.chunk_id === citation.chunk_id) && !filteredCitations.find((c) => c.id === citationIndex) ;
+    return filteredCitations.some((c) => c.chunk_id === citation.chunk_id) ;
 };
 
 export function parseAnswer(answer: AskResponse): ParsedAnswer {

code/tests/functional/tests/backend_api/default/test_conversation.py

@@ -2,6 +2,7 @@
 import re
 import pytest
 from pytest_httpserver import HTTPServer
+from unittest.mock import patch
 import requests
 
 from tests.request_matching import (
@@ -176,7 +177,9 @@ def test_post_makes_correct_calls_to_openai_embeddings_to_embed_question_to_sear
 
 
 def test_post_makes_correct_calls_to_openai_embeddings_to_embed_question_to_store_in_conversation_log(
-    app_url: str, app_config: AppConfig, httpserver: HTTPServer
+    app_url: str,
+    app_config: AppConfig,
+    httpserver: HTTPServer,
 ):
     # when
     requests.post(f"{app_url}{path}", json=body)
@@ -649,9 +652,15 @@ def test_post_makes_correct_call_to_store_conversation_in_search(
     )
 
 
+@patch(
+    "backend.batch.utilities.helpers.config.config_helper.ConfigHelper.get_active_config_or_default"
+)
 def test_post_returns_error_when_downstream_fails(
-    app_url: str, app_config: AppConfig, httpserver: HTTPServer
+    get_active_config_or_default_mock, app_url: str, httpserver: HTTPServer
 ):
+    get_active_config_or_default_mock.return_value.prompts.conversational_flow = (
+        "custom"
+    )
     httpserver.expect_oneshot_request(
         re.compile(".*"),
     ).respond_with_json({}, status=403)