Merge branch 'dev' into dependabotchanges

.env.sample

@@ -22,8 +22,9 @@ AZURE_SEARCH_DATASOURCE_NAME=
 # Azure OpenAI for generating the answer and computing the embedding of the documents
 AZURE_OPENAI_RESOURCE=
 AZURE_OPENAI_API_KEY=
-AZURE_OPENAI_MODEL_INFO="{\"model\":\"gpt-35-turbo-16k\",\"modelName\":\"gpt-35-turbo-16k\",\"modelVersion\":\"0613\"}"
-AZURE_OPENAI_EMBEDDING_MODEL_INFO="{\"model\":\"text-embedding-ada-002\",\"modelName\":\"text-embedding-ada-002\",\"modelVersion\":\"2\"}"
+AZURE_OPENAI_MODEL=gpt-35-turbo
+AZURE_OPENAI_MODEL_NAME=gpt-35-turbo
+AZURE_OPENAI_EMBEDDING_MODEL=text-embedding-ada-002
 AZURE_OPENAI_TEMPERATURE=0
 AZURE_OPENAI_TOP_P=1.0
 AZURE_OPENAI_MAX_TOKENS=1000
@@ -35,10 +36,12 @@ AZURE_OPENAI_STREAM=True
 AzureWebJobsStorage=
 BACKEND_URL=http://localhost:7071
 DOCUMENT_PROCESSING_QUEUE_NAME=
-# Azure Blob Storage for storing the original documents to be processed
-AZURE_BLOB_STORAGE_INFO="{\"containerName\":\"documents\",\"accountName\":\"\",\"accountKey\":\"\"}"
+AZURE_BLOB_ACCOUNT_NAME=
+AZURE_BLOB_ACCOUNT_KEY=
+AZURE_BLOB_CONTAINER_NAME=
 # Azure Form Recognizer for extracting the text from the documents
-AZURE_FORM_RECOGNIZER_INFO="{\"endpoint\":\"\",\"key\":\"\"}"
+AZURE_FORM_RECOGNIZER_ENDPOINT=
+AZURE_FORM_RECOGNIZER_KEY=
 # Azure AI Content Safety for filtering out the inappropriate questions or answers
 AZURE_CONTENT_SAFETY_ENDPOINT=
 AZURE_CONTENT_SAFETY_KEY=
@@ -60,8 +63,11 @@ AZURE_KEY_VAULT_ENDPOINT=
 # Chat conversation type to decide between custom or byod (bring your own data) conversation type
 CONVERSATION_FLOW=
 # Chat History CosmosDB Integration Settings
-AZURE_COSMOSDB_INFO="{\"accountName\":\"cosmos-abc123\",\"databaseName\":\"db_conversation_history\",\"containerName\":\"conversations\"}"
-AZURE_COSMOSDB_ACCOUNT_KEY=
+AZURE_COSMOSDB_ACCOUNT_NAME=
+AZURE_COSMOSDB_DATABASE_NAME=
+AZURE_COSMOSDB_CONVERSATIONS_CONTAINER_NAME=
 AZURE_COSMOSDB_ENABLE_FEEDBACK=
-AZURE_POSTGRESQL_INFO="{\"user\":\"\",\"dbname\":\"postgres\",\"host\":\"\"}"
+AZURE_POSTGRESQL_HOST_NAME=
+AZURE_POSTGRESQL_DATABASE_NAME=
+AZURE_POSTGRESQL_USER=
 DATABASE_TYPE="CosmosDB"

.github/workflows/build-docker-images.yml

@@ -1,9 +1,7 @@
 name: Build Docker Images
 
 on:
-  workflow_run:
-    workflows: [Tests]
-    types: [completed]
+  push:
     branches:
       - main
       - dev
@@ -22,7 +20,6 @@ on:
 
 jobs:
   docker-build:
-    if: ${{ github.event_name != 'workflow_run' || github.event.workflow_run.conclusion == 'success' }}
     strategy:
       matrix:
         include:
@@ -34,9 +31,9 @@ jobs:
             dockerfile: docker/Frontend.Dockerfile
     uses: ./.github/workflows/build-docker.yml
     with:
-      registry: ${{ github.event.workflow_run.head_branch == 'main' && 'fruoccopublic.azurecr.io' || 'cwydcontainerreg.azurecr.io'}}
-      username: ${{ github.event.workflow_run.head_branch == 'main' && 'fruoccopublic' || 'cwydcontainerreg'}}
+      registry: ${{ github.ref_name == 'main' && 'fruoccopublic.azurecr.io' || 'cwydcontainerreg.azurecr.io'}}
+      username: ${{ github.ref_name == 'main' && 'fruoccopublic' || 'cwydcontainerreg'}}
       app_name: ${{ matrix.app_name }}
       dockerfile: ${{ matrix.dockerfile }}
-      push: ${{ github.event.workflow_run.head_branch == 'main' || github.event.workflow_run.head_branch == 'dev' || github.event.workflow_run.head_branch == 'demo' }}
+      push: ${{ github.ref_name == 'main' || github.ref_name == 'dev' || github.ref_name == 'demo' }}
     secrets: inherit

.github/workflows/build-docker.yml

@@ -28,7 +28,6 @@ jobs:
   docker-build:
     runs-on: ubuntu-latest
     steps:
-
     - name: Checkout
       uses: actions/checkout@v4
 
@@ -61,7 +60,7 @@ jobs:
         context: .
         file: ${{ inputs.dockerfile }}
         push: ${{ inputs.push }}
-        cache-from: type=registry,ref=${{ inputs.registry }}/${{ inputs.app_name}}:${{ github.ref_name == 'main' && 'latest' || github.ref_name == 'dev' && 'dev' || github.ref_name == 'demo' && 'demo' || 'latest' }}
+        cache-from: type=registry,ref=${{ inputs.registry }}/${{ inputs.app_name}}:${{ github.ref_name == 'main' && 'latest' || github.ref_name == 'dev' && 'dev' || github.ref_name == 'demo' && 'demo' || github.head_ref }}
         tags: |
-          ${{ inputs.registry }}/${{ inputs.app_name}}:${{ github.ref_name == 'main' && 'latest' || github.ref_name == 'dev' && 'dev' || github.ref_name == 'demo' && 'demo' || 'latest' }}
+          ${{ inputs.registry }}/${{ inputs.app_name}}:${{ github.ref_name == 'main' && 'latest' || github.ref_name == 'dev' && 'dev' || github.ref_name == 'demo' && 'demo' || github.head_ref }}
           ${{ inputs.registry }}/${{ inputs.app_name}}:${{ steps.date.outputs.date }}_${{ github.run_number }}

README.md

@@ -50,6 +50,7 @@ Welcome to the *Chat with your data* Solution accelerator repository! The *Chat
 
 
 
+
 ### About this repo
 
 This repository provides an end-to-end solution for users who want to query their data with natural language. It includes a well designed ingestion mechanism for multiple file types, an easy deployment, and a support team for maintenance. The accelerator demonstrates both Push or Pull Ingestion; the choice of orchestration (Semantic Kernel, LangChain, OpenAI Functions or [Prompt Flow](docs/prompt_flow.md)) and should be the minimum components needed to implement a RAG pattern. It is not intended to be put into Production as-is without experimentation or evaluation of your data. It provides the following features:

azure.yaml

@@ -6,6 +6,12 @@ metadata:
 hooks:
   postprovision:
     run: ./infra/prompt-flow/create-prompt-flow.sh
+    posix:
+      shell: sh
+      run: ./scripts/parse_env.sh
+    windows:
+      shell: pwsh
+      run: ./scripts/parse_env.ps1
 services:
   web:
     project: ./code

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

@@ -247,7 +247,7 @@ def get_container_sas(self):
             user_delegation_key=self.user_delegation_key,
             account_key=self.account_key,
             permission="r",
-            expiry=datetime.utcnow() + timedelta(hours=1),
+            expiry=datetime.utcnow() + timedelta(days=365 * 5),
         )
 
     def get_blob_sas(self, file_name):

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

@@ -52,6 +52,9 @@ def __init__(self, config: dict):
         )
         self.enable_chat_history = config["enable_chat_history"]
         self.database_type = config.get("database_type", self.env_helper.DATABASE_TYPE)
+        self.conversational_flow = config.get(
+            "conversational_flow", self.env_helper.CONVERSATION_FLOW
+        )
 
     def get_available_document_types(self) -> list[str]:
         document_types = {
@@ -247,11 +250,7 @@ def get_default_config():
                 logger.info("Loading default config from %s", config_file_path)
                 ConfigHelper._default_config = json.loads(
                     Template(f.read()).substitute(
-                        ORCHESTRATION_STRATEGY=(
-                            OrchestrationStrategy.SEMANTIC_KERNEL.value
-                            if env_helper.DATABASE_TYPE == DatabaseType.POSTGRESQL.value
-                            else env_helper.ORCHESTRATION_STRATEGY
-                        ),
+                        ORCHESTRATION_STRATEGY=env_helper.ORCHESTRATION_STRATEGY,
                         LOG_USER_INTERACTIONS=(
                             False
                             if env_helper.DATABASE_TYPE == DatabaseType.POSTGRESQL.value
@@ -262,6 +261,7 @@ def get_default_config():
                             if env_helper.DATABASE_TYPE == DatabaseType.POSTGRESQL.value
                             else True
                         ),
+                        CONVERSATION_FLOW=env_helper.CONVERSATION_FLOW,
                         DATABASE_TYPE=env_helper.DATABASE_TYPE,
                     )
                 )

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

@@ -9,7 +9,7 @@
     "enable_post_answering_prompt": false,
     "ai_assistant_type": "default",
     "enable_content_safety": true,
-    "conversational_flow": "custom"
+    "conversational_flow": "${CONVERSATION_FLOW}"
   },
   "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

@@ -5,6 +5,9 @@
 from dotenv import load_dotenv
 from azure.identity import DefaultAzureCredential, get_bearer_token_provider
 from azure.keyvault.secrets import SecretClient
+
+from ..orchestrator.orchestration_strategy import OrchestrationStrategy
+from ..helpers.config.conversation_flow import ConversationFlow
 from ..helpers.config.database_type import DatabaseType
 
 logger = logging.getLogger(__name__)
@@ -97,11 +100,24 @@ def __load_config(self, **kwargs) -> None:
         # Cosmos DB configuration
         if self.DATABASE_TYPE == DatabaseType.COSMOSDB.value:
             azure_cosmosdb_info = self.get_info_from_env("AZURE_COSMOSDB_INFO", "")
-            self.AZURE_COSMOSDB_DATABASE = azure_cosmosdb_info.get("databaseName", "")
-            self.AZURE_COSMOSDB_ACCOUNT = azure_cosmosdb_info.get("accountName", "")
-            self.AZURE_COSMOSDB_CONVERSATIONS_CONTAINER = azure_cosmosdb_info.get(
-                "containerName", ""
-            )
+            if azure_cosmosdb_info:
+                self.AZURE_COSMOSDB_DATABASE = azure_cosmosdb_info.get(
+                    "databaseName", ""
+                )
+                self.AZURE_COSMOSDB_ACCOUNT = azure_cosmosdb_info.get("accountName", "")
+                self.AZURE_COSMOSDB_CONVERSATIONS_CONTAINER = azure_cosmosdb_info.get(
+                    "containerName", ""
+                )
+            else:
+                self.AZURE_COSMOSDB_DATABASE = os.getenv(
+                    "AZURE_COSMOSDB_DATABASE_NAME", ""
+                )
+                self.AZURE_COSMOSDB_ACCOUNT = os.getenv(
+                    "AZURE_COSMOSDB_ACCOUNT_NAME", ""
+                )
+                self.AZURE_COSMOSDB_CONVERSATIONS_CONTAINER = os.getenv(
+                    "AZURE_COSMOSDB_CONVERSATIONS_CONTAINER_NAME", ""
+                )
             self.AZURE_COSMOSDB_ACCOUNT_KEY = self.secretHelper.get_secret(
                 "AZURE_COSMOSDB_ACCOUNT_KEY"
             )
@@ -114,18 +130,32 @@ def __load_config(self, **kwargs) -> None:
             self.USE_ADVANCED_IMAGE_PROCESSING = self.get_env_var_bool(
                 "USE_ADVANCED_IMAGE_PROCESSING", "False"
             )
+            self.CONVERSATION_FLOW = os.getenv("CONVERSATION_FLOW", "custom")
+            # Orchestration Settings
+            self.ORCHESTRATION_STRATEGY = os.getenv(
+                "ORCHESTRATION_STRATEGY", "openai_function"
+            )
         # PostgreSQL configuration
         elif self.DATABASE_TYPE == DatabaseType.POSTGRESQL.value:
             self.AZURE_POSTGRES_SEARCH_TOP_K = self.get_env_var_int(
                 "AZURE_POSTGRES_SEARCH_TOP_K", 5
             )
             azure_postgresql_info = self.get_info_from_env("AZURE_POSTGRESQL_INFO", "")
-            self.POSTGRESQL_USER = azure_postgresql_info.get("user", "")
-            self.POSTGRESQL_DATABASE = azure_postgresql_info.get("dbname", "")
-            self.POSTGRESQL_HOST = azure_postgresql_info.get("host", "")
+            if azure_postgresql_info:
+                self.POSTGRESQL_USER = azure_postgresql_info.get("user", "")
+                self.POSTGRESQL_DATABASE = azure_postgresql_info.get("dbname", "")
+                self.POSTGRESQL_HOST = azure_postgresql_info.get("host", "")
+            else:
+                self.POSTGRESQL_USER = os.getenv("AZURE_POSTGRESQL_USER", "")
+                self.POSTGRESQL_DATABASE = os.getenv(
+                    "AZURE_POSTGRESQL_DATABASE_NAME", ""
+                )
+                self.POSTGRESQL_HOST = os.getenv("AZURE_POSTGRESQL_HOST_NAME", "")
             # Ensure integrated vectorization is disabled for PostgreSQL
             self.AZURE_SEARCH_USE_INTEGRATED_VECTORIZATION = False
             self.USE_ADVANCED_IMAGE_PROCESSING = False
+            self.CONVERSATION_FLOW = ConversationFlow.CUSTOM.value
+            self.ORCHESTRATION_STRATEGY = OrchestrationStrategy.SEMANTIC_KERNEL.value
         else:
             raise ValueError(
                 "Unsupported DATABASE_TYPE. Please set DATABASE_TYPE to 'CosmosDB' or 'PostgreSQL'."
@@ -305,10 +335,6 @@ def __load_config(self, **kwargs) -> None:
         self.AZURE_CONTENT_SAFETY_KEY = self.secretHelper.get_secret(
             "AZURE_CONTENT_SAFETY_KEY"
         )
-        # Orchestration Settings
-        self.ORCHESTRATION_STRATEGY = os.getenv(
-            "ORCHESTRATION_STRATEGY", "openai_function"
-        )
         # 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/frontend/src/pages/chat/Chat.tsx

@@ -69,6 +69,7 @@ const Chat = () => {
   const lastQuestionRef = useRef<string>("");
   const chatMessageStreamEnd = useRef<HTMLDivElement | null>(null);
   const [isLoading, setIsLoading] = useState<boolean>(false);
+  const [isGenerating, setIsGenerating] = useState<boolean>(false);  // Add this state
   const [showLoadingMessage, setShowLoadingMessage] = useState<boolean>(false);
   const [isAssistantAPILoading, setIsAssistantAPILoading] = useState(false);
   const [isSendButtonDisabled, setSendButtonDisabled] = useState(false);
@@ -185,7 +186,7 @@ const Chat = () => {
       text: "Clear all chat history",
       disabled:
         !chatHistory.length ||
-        isLoading ||
+        isGenerating ||
         fetchingConvMessages ||
         fetchingChatHistory,
       iconProps: { iconName: "Delete" },
@@ -194,7 +195,7 @@ const Chat = () => {
   const makeApiRequest = async (question: string) => {
     lastQuestionRef.current = question;
 
-    setIsLoading(true);
+    setIsGenerating(true);
     setShowLoadingMessage(true);
     const abortController = new AbortController();
     abortFuncs.current.unshift(abortController);
@@ -274,7 +275,7 @@ const Chat = () => {
       }
       setAnswers([...answers, userMessage]);
     } finally {
-      setIsLoading(false);
+      setIsGenerating(false);
       setShowLoadingMessage(false);
       abortFuncs.current = abortFuncs.current.filter(
         (a) => a !== abortController
@@ -371,7 +372,7 @@ const Chat = () => {
   const stopGenerating = () => {
     abortFuncs.current.forEach((a) => a.abort());
     setShowLoadingMessage(false);
-    setIsLoading(false);
+    setIsGenerating(false);
   };
 
   useEffect(() => {
@@ -485,6 +486,10 @@ const Chat = () => {
   };
 
   const onSelectConversation = async (id: string) => {
+    if (isGenerating) {
+      // If response is being generated, prevent switching threads
+      return;
+    }
     if (!id) {
       console.error("No conversation Id found");
       return;
@@ -623,7 +628,7 @@ const Chat = () => {
             ) : (
               <div
                 className={styles.chatMessageStream}
-                style={{ marginBottom: isLoading ? "40px" : "0px" }}
+                style={{ marginBottom: isGenerating ? "40px" : "0px" }}
               >
                 {fetchingConvMessages && (
                   <div className={styles.fetchMessagesSpinner}>
@@ -697,7 +702,7 @@ const Chat = () => {
             </div>
 
             <Stack horizontal className={styles.chatInput}>
-              {isLoading && (
+              {isGenerating && (
                 <Stack
                   horizontal
                   className={styles.stopGeneratingContainer}
@@ -725,10 +730,10 @@ const Chat = () => {
                 className={`${styles.clearChatBroom} ${styles.mobileclearChatBroom}`}
                 style={{
                   background:
-                    isLoading || answers.length === 0
+                    isGenerating || answers.length === 0
                       ? "#BDBDBD"
                       : "radial-gradient(109.81% 107.82% at 100.1% 90.19%, #0F6CBD 33.63%, #2D87C3 70.31%, #8DDDD8 100%)",
-                  cursor: isLoading || answers.length === 0 ? "" : "pointer",
+                  cursor: isGenerating || answers.length === 0 ? "" : "pointer",
                 }}
                 onClick={clearChat}
                 onKeyDown={(e) =>
@@ -741,7 +746,7 @@ const Chat = () => {
               <QuestionInput
                 clearOnSend
                 placeholder="Type a new question..."
-                disabled={isLoading}
+                disabled={isGenerating}
                 onSend={(question) => makeApiRequest(question)}
                 recognizedText={recognizedText}
                 isSendButtonDisabled={isSendButtonDisabled}