Mixtral 8x7B

app.py

@@ -6,6 +6,7 @@ import datetime
 from langchain.tools import tool
 from langchain_huggingface import HuggingFaceEndpoint, ChatHuggingFace
 from langchain.agents import initialize_agent, AgentType
+from bs4 import BeautifulSoup
 
 ## # Load environment variables from .env file
 # --- Constants ---
@@ -195,6 +196,27 @@ def classify_image(image_url: str) -> str:
     except Exception:
         return "error"
 
+# --- TOOL 12: Web Scraping Tool ---
+@tool
+def web_scrape_tool(url: str) -> str:
+    """
+    Scrape the main textual content from a given website URL and return a concise summary or answer.
+    Input: A valid URL (e.g., 'https://en.wikipedia.org/wiki/Python_(programming_language)')
+    """
+    try:
+        headers = {
+            "User-Agent": "Mozilla/5.0 (compatible; WebScrapeTool/1.0)"
+        }
+        resp = requests.get(url, headers=headers, timeout=20)
+        resp.raise_for_status()
+        soup = BeautifulSoup(resp.text, "html.parser")
+        # Try to extract main content from common tags
+        paragraphs = soup.find_all("p")
+        text = " ".join(p.get_text() for p in paragraphs)
+        # Limit to first 1000 characters for brevity
+        return text[:1000] if text else "No textual content found."
+    except Exception as e:
+        return f"error: {e}"
 
 ##-- Tool Discovery ---
 # Use @tool for each function.
@@ -212,7 +234,8 @@ tools_list = [
     currency_convert,
     image_caption,
     ocr_image,
-    classify_image
+    classify_image,
+    web_scrape_tool
 ]
 
 tool_descriptions = "\n".join(f"- {tool.name}: {tool.description}" for tool in tools_list)
@@ -227,6 +250,8 @@ You are an intelligent assistant with access to the following tools:
 
 For every question, you must do your internal reasoning using the Thought → Action → Observation → Answer process, but your output to the user should be ONLY the final answer as a single value (number, string, or comma-separated list), with no extra explanation, thoughts, actions, or observations.
 
+**If a tool returns a long text or description (such as from a web scraping tool), you must carefully read and process that output, and extract or identify ONLY the most relevant, concise answer to the user's question, and provide a single string as output. Do not return the full text or irrelevant details.**
+
 **Your output must be only the answer. Do not include any reasoning, tool calls, or explanations.**
 
 Examples:
@@ -240,6 +265,10 @@ Your Output: 22
 Q: What is the capital of France?
 Your Output: Paris
 
+Q: Which year was python 3.0 released as per the website https://en.wikipedia.org/wiki/Python_(programming_language)?
+(Tool returns a long description about Python.)
+Your Output: 2008
+
 Q: Convert 10 meters to feet.
 Your Output: 32.81
 
@@ -247,6 +276,7 @@ Instructions:
 - Always do your internal reasoning (Thought → Action → Observation → Answer) before producing the answer, but DO NOT show this reasoning to the user.
 - Use a tool only if necessary, and don't use multiple tools in a call. Don't use a tool if you can answer directly.
 - Your output must be a single value (number, string, or comma-separated list) with no extra explanation or formatting.
+- If you cannot answer the question or if you couldn't process the input question just answer as "no_answer".
 - Be concise and accurate.
 """
 
@@ -254,7 +284,8 @@ Instructions:
 # Generate the chat interface, including the tools
 
 llm = HuggingFaceEndpoint(
-    repo_id="Qwen/Qwen2.5-32B-Instruct",
+    repo_id="mistralai/Mixtral-8x7B-Instruct-v0.1",
+    # repo_id="Qwen/Qwen2.5-32B-Instruct",
     huggingfacehub_api_token=HF_ACCESS_KEY,
     # model_kwargs={'prompt': system_prompt}
     # system_prompt=system_prompt,

requirements.txt

@@ -7,4 +7,5 @@ huggingface-hub
 langchain-huggingface
 langchain-community
 transformers
-langchain-openai
+langchain-openai
+beautifulsoup4