Add final formatting step

app.py

@@ -3,7 +3,8 @@ import gradio as gr
 import requests
 import inspect
 import pandas as pd
-from llama_index.core.agent.workflow import AgentWorkflow
+from llama_index.core.workflow import Context
+from llama_index.core.agent.workflow import AgentWorkflow, ToolCallResult, AgentStream
 from llama_index.llms.huggingface_api import HuggingFaceInferenceAPI
 from tools import (
     APIProcessor,
@@ -11,8 +12,9 @@ from tools import (
     transcribe_image_from_link,
     transcribe_webpage,
 )
+from utils import format_final_answer
 from search import GoogleSearch
-from dotenv import load_dotenv
+import asyncio
 
 # (Keep Constants as is)
 # --- Constants ---
@@ -30,13 +32,13 @@ class BasicAgent:
         self.llm = HuggingFaceInferenceAPI(model_name="Qwen/Qwen2.5-Coder-32B-Instruct")
         print("BasicAgent initialized.")
 
-    def __call__(self, question: str, task_id: str, file_name: str) -> str:
+    async def __call__(self, question: str, task_id: str, file_name: str) -> str:
         google_search = GoogleSearch().google_search
         google_image_search = GoogleSearch().google_image_search
 
         get_and_process_question_attachment = APIProcessor(
             file_url=DEFAULT_API_URL + "/files/" + task_id, file_name=file_name
-        ).get_and_process_attachment()
+        ).get_and_process_attachment
 
         agent = AgentWorkflow.from_tools_or_functions(
             [
@@ -50,10 +52,22 @@ class BasicAgent:
             llm=self.llm,
             system_prompt=SYSTEM_PROMPT,
         )
+        ctx = Context(agent)
+        handler = agent.run(question, ctx=ctx)
+        async for ev in handler.stream_events():
+            if isinstance(ev, ToolCallResult):
+                print("")
+                print(
+                    "Called tool: ", ev.tool_name, ev.tool_kwargs, "=>", ev.tool_output
+                )
+            elif isinstance(ev, AgentStream):  # showing the thought process
+                print(ev.delta, end="", flush=True)
 
-        response = agent.run(question)
+        response = await handler
 
-        return response
+        ##format final answer
+        final_answer = format_final_answer(question, response)
+        return final_answer
 
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
@@ -226,6 +240,36 @@ with gr.Blocks() as demo:
 
     run_button.click(fn=run_and_submit_all, outputs=[status_output, results_table])
 
+
+# async def main():
+#     agent = BasicAgent()
+#     api_url = DEFAULT_API_URL
+#     questions_url = f"{api_url}/questions"
+#     print(f"Fetching questions from: {questions_url}")
+
+#     response = requests.get(questions_url, timeout=15)
+#     response.raise_for_status()
+#     questions_data = response.json()
+
+#     # 3. Run your Agent
+#     results_log = []
+#     answers_payload = []
+#     print(f"Running agent on {len(questions_data)} questions...")
+#     item = questions_data[0]
+#     task_id = item.get("task_id")
+#     question_text = item.get("question")
+#     file_name = item.get("file_name")
+#     submitted_answer = await agent(question_text, task_id, file_name)
+#     answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
+#     results_log.append(
+#         {
+#             "Task ID": task_id,
+#             "Question": question_text,
+#             "Submitted Answer": submitted_answer,
+#         }
+#     )
+
+
 if __name__ == "__main__":
     print("\n" + "-" * 30 + " App Starting " + "-" * 30)
     # Check for SPACE_HOST and SPACE_ID at startup for information
@@ -253,3 +297,5 @@ if __name__ == "__main__":
 
     print("Launching Gradio Interface for Basic Agent Evaluation...")
     demo.launch(debug=True, share=False)
+
+    # asyncio.run(main())

requirements.txt

@@ -7,6 +7,6 @@ pandas
 aiohttp
 Pillow
 yt-dlp
-markdownify
+html2text
 llama-index-utils-workflow
 llama-index-llms-huggingface-api

tools.py

@@ -8,8 +8,9 @@ import os
 import io
 import yt_dlp
 import re
-from markdownify import markdownify
+import html2text
 from requests.exceptions import RequestException
+from bs4 import BeautifulSoup
 
 
 def transcribe_image_from_link(image_link: str) -> str:
@@ -49,6 +50,13 @@ def transcribe_image_from_link(image_link: str) -> str:
     return transcribed_text
 
 
+def truncate_content(content: str, max_length: int = 10000) -> str:
+    if len(content) <= max_length:
+        return content
+    else:
+        return content[:max_length]
+
+
 def transcribe_webpage(website_url: str) -> str:
     """Visits website url and returns markdown of contents"""
     try:
@@ -56,21 +64,25 @@ def transcribe_webpage(website_url: str) -> str:
         response = requests.get(website_url, timeout=20)
         response.raise_for_status()  # Raise an exception for bad status codes
 
-        # Convert the HTML content to Markdown
-        markdown_content = markdownify(response.text).strip()
+        soup = BeautifulSoup(response.text, "html.parser")
+        content_div = soup.find("div", id="mw-content-text")
+
+        if not content_div:
+            return "Main content not found."
+
+        # Only extract <p> and <table> tags
+        elements = content_div.find_all(["p", "table"])
 
-        with open("webpage_content.md", "a") as f:
-            f.write("""Response from the webpage:\n\n""")
-            f.write(response.text)
+        # Join selected HTML chunks
+        html_subset = "".join(str(el) for el in elements)
+
+        # Convert the HTML content to Markdown
+        markdown_content = html2text.HTML2Text().handle(str(html_subset))
 
         # Remove multiple line breaks
         markdown_content = re.sub(r"\n{3,}", "\n\n", markdown_content)
 
-        with open("webpage_content.md", "a") as f:
-            f.write("\n\nMarkdown content:\n\n")
-            f.write(markdown_content)
-
-        return markdown_content
+        return truncate_content(markdown_content, 20000)
 
     except requests.exceptions.Timeout:
         return "The request timed out. Please try again later or check the URL."
@@ -218,11 +230,15 @@ if __name__ == "__main__":
     #     return "https://agents-course-unit4-scoring.hf.space" + "/files/" + task_id
 
     # audio_task_processor = APIProcessor(
-    #     file_name="7bd855d8-463d-4ed5-93ca-5fe35145f733.xlsx",
-    #     file_url=get_file_api_url("7bd855d8-463d-4ed5-93ca-5fe35145f733"),
+    #     file_name="",
+    #     file_url=get_file_api_url("8e867cd7-cff9-4e6c-867a-ff5ddc2550be"),
     # )
 
     # response = audio_task_processor.get_and_process_attachment()
     # print(response)
-    result = parse_youtube_video("https://www.youtube.com/watch?v=1htKBjuUWec")
-    print(result)
+    # result = parse_youtube_video("https://www.youtube.com/watch?v=1htKBjuUWec")
+    # print(result)
+    text = transcribe_webpage(
+        "https://en.wikipedia.org/wiki/Mercedes_Sosa#Studio_albums"
+    )
+    print(text)

utils.py

@@ -1,6 +1,7 @@
 from PIL import Image
 from io import BytesIO
 import base64
+from openai import OpenAI
 
 
 def encode_image_in_base64(image: bytes):
@@ -50,3 +51,27 @@ def process_image_for_gpt(image_bytes: bytes) -> str:
     image_bytes = replace_transparent_pixels(image_bytes)
     base64_image = encode_image_in_base64(image_bytes)
     return base64_image
+
+
+def format_final_answer(question: str, answer: str) -> str:
+    """Always call to format final answer"""
+    client = OpenAI()
+    response = client.chat.completions.create(
+        model="gpt-4o",
+        messages=[
+            {
+                "role": "system",
+                "content": """ You're tasked with reformatting an answer from an unreliable AI into the expected format as per their instructions.
+            <instructions>
+            You are a general AI assistant. I will ask you a question. Report your thoughts, and finish your answer with the following template: FINAL ANSWER: [YOUR FINAL ANSWER]. YOUR FINAL ANSWER should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
+            </instructions>
+            <question>"""
+                + question
+                + """
+            Now here is their answer. Only reply with the corrected formatting
+            """,
+            },
+            {"role": "user", "content": str(answer)},
+        ],
+    )
+    return response.choices[0].message.content.strip()