fix: issues with requirements

.github/workflows/azure-dev.yml

@@ -48,7 +48,7 @@ jobs:
         uses: actions/checkout@v4
 
       - name: Install azd
-        uses: Azure/setup-azd@v1.0.0
+        uses: Azure/setup-azd@v2
 
       - name: Install Nodejs
         uses: actions/setup-node@v4

.github/workflows/evaluate.yml

@@ -31,13 +31,15 @@ jobs:
       AZURE_OPENAI_NAME: ${{ vars.AZURE_OPENAI_NAME }}
       BING_SEARCH_ENDPOINT: ${{ vars.BING_SEARCH_ENDPOINT }}
       BING_SEARCH_KEY: ${{ secrets.BING_SEARCH_KEY }}
+      AZURE_AI_PROJECT_NAME: ${{ vars.AZURE_AI_PROJECT_NAME }}
+      AZURE_LOCATION: ${{ vars.AZURE_LOCATION }}
 
     steps:
       - name: checkout repo content
         uses: actions/checkout@v4 # checkout the repository content
 
       - name: Install azd
-        uses: Azure/setup-azd@v1.0.0
+        uses: Azure/setup-azd@v2
 
       - name: setup python
         uses: actions/setup-python@v5

azure.yaml

@@ -39,11 +39,12 @@ pipeline:
     - AZURE_SEARCH_ENDPOINT
     - BING_SEARCH_ENDPOINT
     - AZURE_OPENAI_NAME
-    - AZURE_CONTAINER_REGISTRY_NAME   
-    - AZURE_RESOURCE_GROUP
+    - AZURE_CONTAINER_REGISTRY_NAME
     - API_SERVICE_ACA_NAME
     - WEB_SERVICE_ACA_NAME
     - AZURE_SEARCH_NAME
     - WEB_SERVICE_ACA_URI
+    - AZURE_AI_PROJECT_NAME
+    - AZURE_LOCATION
   secrets:
     - BING_SEARCH_KEY

docs/workshop/LAB-MANUAL.md

@@ -29,7 +29,20 @@ To participate in this workshop, you will need:
 
 4. Once your Codespace is ready, **run the following command**:
 
-++./docs/workshop/lab_setup.py --username '@lab.CloudPortalCredential(User1).Username' --password '@lab.CloudPortalCredential(User1).Password' --azure-env-name 'AITOUR@lab.LabInstance.Id' --subscription '@lab.CloudSubscription.Id'++
+```bash
+# 1. Install the required dependencies
+pip install -r requirements.txt
+
+# 2.a. Run the following script if you haven't already provsionned the environment with "azd up"
+./docs/workshop/lab_setup.py --username '@lab.CloudPortalCredential(User1).Username' --password '@lab.CloudPortalCredential(User1).Password' --azure-env-name '<your-env-name>' --subscription '@lab.CloudSubscription.Id'
+
+# 2.b. Else run the following command if you have already run "azd up":
+azd env get-values > .env 
+
+# 3. In all cases, you need to run the following commands:
+sed 's/^/export /' .env >> ~/.bashrc # export variables to your shell
+source ~/.bashrc # refresh your shell
+```
 
 
 > [!IMPORTANT]

docs/workshop/researcher/researcher-0.prompty

@@ -11,6 +11,12 @@ model:
     type: azure_openai
     azure_deployment: gpt-4
     api_version: 2024-08-01-preview
+  parameters:
+    temperature: 0.2
+    max_tokens: 1000
+    top_p: 1
+    frequency_penalty: 0
+    presence_penalty: 0
 sample:
   instructions: Can you generate queries to find the latest winter camping trends? Use 'en-US' as the market code. 
 ---

docs/workshop/researcher/researcher-1.prompty

@@ -11,6 +11,12 @@ model:
     type: azure_openai
     azure_deployment: gpt-4
     api_version: 2024-08-01-preview
+  parameters:
+    temperature: 0.2
+    max_tokens: 1000
+    top_p: 1
+    frequency_penalty: 0
+    presence_penalty: 0
 inputs:
   participant:
     type: object

docs/workshop/researcher/researcher3.py

@@ -1,86 +1,42 @@
 import os
 import json
+import sys
+sys.path.append(os.path.abspath('../../../src/api/agents/researcher'))
+from researcher import execute_research
 from typing import List
-import requests
-import urllib.parse
 from dotenv import load_dotenv
 import prompty
 import prompty.azure
 from prompty.azure.processor import ToolCall
 
 load_dotenv()
 
-BING_SEARCH_ENDPOINT = os.getenv("BING_SEARCH_ENDPOINT")
-BING_SEARCH_KEY = os.getenv("BING_SEARCH_KEY")
-BING_HEADERS = {"Ocp-Apim-Subscription-Key": BING_SEARCH_KEY}
+global_prompty_file_path=os.getcwd()+'/researcher-2.prompty'
 
+def find_information(query, market='en-US'):
+    # You can use another api to find information, here we are using "Grounding with Bing Search"
+    print("Executing 'Find Information' ToolCall in %s.." % market)
+    return execute_research(instructions=query, return_raw=True, prompty_file_path=global_prompty_file_path)
 
-def _make_bing_endpoint(endpoint, path):
-    """Make a Bing endpoint URL"""
-    return f"{endpoint}{'' if endpoint.endswith('/') else '/'}{path}"
+def find_entities(query, market='en-US'):
+    # You can use another api to find entities, here we are using "Grounding with Bing Search"
+    print("Executing 'Find Entities' ToolCall in %s.." % market)
+    return execute_research(instructions=query, return_raw=True, prompty_file_path=global_prompty_file_path)
 
+def find_news(query, market='en-US'):
+    print("Executing 'Find News' ToolCall in %s.." % market)
+    # You can use another api to find bews, here we are using "Grounding with Bing Search"
+    return execute_research(instructions=query, return_raw=True, prompty_file_path=global_prompty_file_path)
 
-def _make_bing_request(path, params=None):
-    """Make a request to the Bing API"""
-    endpoint = _make_bing_endpoint(BING_SEARCH_ENDPOINT, path)
-    response = requests.get(endpoint, headers=BING_HEADERS, params=params)
-    items = response.json()
-    return items
-
-
-def find_information(query, market="en-US"):
-    """Find information using the Bing Search API"""
-    params = {"q": query, "mkt": market, "count": 5}
-    items = _make_bing_request("v7.0/search", params)
-    pages = [
-        {"url": a["url"], "name": a["name"], "description": a["snippet"]}
-        for a in items["webPages"]["value"]
-    ]
-    related = [a["text"] for a in items["relatedSearches"]["value"]]
-    return {"pages": pages, "related": related}
-
-
-
-def find_entities(query, market="en-US"):
-    """Find entities using the Bing Entity Search API"""
-    params = "?mkt=" + market + "&q=" + urllib.parse.quote(query)
-    items = _make_bing_request(f"v7.0/entities{params}")
-    entities = []
-    if "entities" in items:
-        entities = [
-            {"name": e["name"], "description": e["description"]}
-            for e in items["entities"]["value"]
-        ]
-    return entities
-
-
-
-def find_news(query, market="en-US"):
-    """Find news using the Bing News Search API"""
-    params = {"q": query, "mkt": market, "count": 5}
-    items = _make_bing_request("v7.0/news/search", params)
-    articles = [
-        {
-            "name": a["name"],
-            "url": a["url"],
-            "description": a["description"],
-            "provider": a["provider"][0]["name"],
-            "datePublished": a["datePublished"],
-        }
-        for a in items["value"]
-    ]
-    return articles
-
-
-def execute_researcher_prompty(instructions: str):
+def execute_researcher_prompty(instructions: str, prompty_file_path=global_prompty_file_path):
     """
     Executes the researcher prompty to find information, entities, and news,
     and runs the selected function given the query and returns the results
     """
 
     # Execute the researcher prompty
     function_calls: List[ToolCall] = prompty.execute(
-        "researcher-2.prompty", inputs={"instructions": instructions}
+        prompty_file_path, inputs={"instructions": instructions}
     )
 
     return function_calls
@@ -152,13 +108,16 @@ def extract_findings(research):
     }
 
 
-def research(instructions: str):
+def research(instructions: str, prompty_file_path: str = None):
     """
     Calls the execute and process functions above to run the research agent 
     and return the results to the user in a readable format. 
     """
-
-    function_calls = execute_researcher_prompty(instructions=instructions)
+
+    global global_prompty_file_path
+    if prompty_file_path is not None:
+        global_prompty_file_path=prompty_file_path
+    function_calls = execute_researcher_prompty(instructions=instructions, prompty_file_path=global_prompty_file_path)
     research = execute_function_calls(function_calls)
-    findings = extract_findings(research)
-    return findings
+    #findings = extract_findings(research)
+    return research

docs/workshop/socialmedia/social.py

@@ -5,7 +5,7 @@
 from prompty.tracer import trace, Tracer, PromptyTracer
 
 # Import the researcher agent to use here
-sys.path.append(os.path.abspath('../../docs/workshop/researcher/'))
+sys.path.append(os.path.abspath('../researcher/'))
 from researcher3 import research
 
 #initiate local prompty tracing
@@ -22,13 +22,13 @@ def execute_social_media_writer_prompty(research_context: str, research, social_
 
     return reseponse
 
-def run_social_media_agent(instructions: str, social_media_instructions: str):
+def run_social_media_agent(instructions: str, social_media_instructions: str, research_prompty_file_path='../researcher/researcher-2.prompty'):
     """
     Run the social media agent
 
     Execute the researcher prompty to find information, entities, and news
     execute the social media writer prompty to create the twitter thread
     """
-    research_results = research(instructions)
+    research_results = research(instructions, prompty_file_path=os.path.abspath(research_prompty_file_path))
     thread = execute_social_media_writer_prompty(research_context= instructions, research=research_results, social_media_instructions = social_media_instructions)
     print(thread)

requirements.txt

@@ -1,2 +1,3 @@
 click
 rich-click
+marshmallow==3.26.1

src/api/agents/researcher/researcher.py

@@ -21,7 +21,7 @@
 
 
 @trace
-def execute_research(instructions: str, feedback: str = "No feedback"):
+def execute_research(instructions: str, prompty_file_path='researcher.prompty', return_raw=False, feedback: str = "No feedback"):
 
     ai_project_conn_str = os.getenv("AZURE_LOCATION")+".api.azureml.ms;"+os.getenv("AZURE_SUBSCRIPTION_ID")+";"+os.getenv("AZURE_RESOURCE_GROUP")+";"+os.getenv("AZURE_AI_PROJECT_NAME")
 
@@ -30,8 +30,7 @@ def execute_research(instructions: str, feedback: str = "No feedback"):
         conn_str=ai_project_conn_str,
     )
 
-    prompt_template = PromptTemplate.from_prompty(file_path="researcher.prompty")
-
+    prompt_template = PromptTemplate.from_prompty(file_path=prompty_file_path)
 
     instructions = instructions
     feedback= feedback
@@ -45,7 +44,7 @@ def execute_research(instructions: str, feedback: str = "No feedback"):
     # Initialize agent bing tool and add the connection id
     bing = BingGroundingTool(connection_id=conn_id)
 
-    prompt_template = PromptTemplate.from_prompty(file_path="researcher.prompty")
+    #prompt_template = PromptTemplate.from_prompty(file_path="researcher.prompty")
 
     # Create agent with the bing tool and process assistant run
     with project_client:
@@ -106,6 +105,8 @@ def run_agent():
         messages = project_client.agents.list_messages(thread_id=thread.id)
         # print(f"Messages: {messages}")
         research_response = messages.data[0]['content'][0]['text']['value']
+        if return_raw:
+            return research_response
         try: 
             json_r = json.loads(research_response)
         except:

src/api/evaluate/evaluate.py

@@ -136,9 +136,8 @@ def run_orchestrator(research_context, product_context, assignment_context):
     }
 
 @trace
-def evaluate_orchestrator(model_config, project_scope,  data_path):
+def evaluate_orchestrator(model_config, project_scope, data_path):
     writer_evaluator = ArticleEvaluator(model_config, project_scope)
-
     data = []    
     eval_data = []
     print(f"\n===== Creating articles to evaluate using data provided in {data_path}")
@@ -150,17 +149,33 @@ def evaluate_orchestrator(model_config, project_scope,  data_path):
             print(f"generating article {num +1}")
             eval_data.append(run_orchestrator(row["research_context"], row["product_context"], row["assignment_context"]))
 
-    # write out eval data to a file so we can re-run evaluation on it
+    # Write out eval data to a file so we can re-run evaluation on it
     with jsonlines.open(folder + '/eval_data.jsonl', 'w') as writer:
         for row in eval_data:
             writer.write(row)
 
     eval_data_path = folder + '/eval_data.jsonl'
-
     print(f"\n===== Evaluating the generated articles")
-    eval_results = writer_evaluator(data_path=eval_data_path)
-    import pandas as pd
 
+    retries = 0
+    max_retries = 5
+    while retries < max_retries:
+        try:
+            eval_results = writer_evaluator(data_path=eval_data_path)
+            break
+        except Exception as e:
+            if 'rate_limit_exceeded' in str(e):
+                wait_time = (2 ** retries) + random.uniform(0, 1)
+                print(f"Rate limit exceeded. Retrying in {wait_time} seconds...")
+                time.sleep(wait_time)
+                retries += 1
+            else:
+                raise e
+
+    if retries == max_retries:
+        raise Exception("Max retries reached. Exiting...")
+
+    import pandas as pd
     print("Evaluation summary:\n")
     print("View in Azure AI Studio at: " + str(eval_results['studio_url']))
     metrics = {key: [value] for key, value in eval_results['metrics'].items()}
@@ -456,4 +471,4 @@ def make_image_message(url_path):
     # eval_image_result = evaluate_image(project_scope, img_paths)
 
     end=time.time()
-    print(f"Finished evaluate in {end - start}s")
+    print(f"Finished evaluate in {end - start}s")