import os import gradio as gr import requests import inspect import pandas as pd import json from duckduckgo_search import DDGS import ast # For safely evaluating literal structures if needed, though JSON is preferred # --- Constants --- DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space" HF_API_URL = "https://api-inference.huggingface.co/models/deepseek-ai/DeepSeek-V3" TEST_AGENT_KEY = os.getenv("TEST_AGENT_KEY") # Make sure you add this secret in your Hugging Face Space MODEL_NAME = "deepseek-ai/Deepseek-V3" # --- Basic Agent Definition --- class BasicAgent: def __init__(self): print("BasicAgent initialized.") self.api_key = os.getenv("TEST_AGENT_KEY") if not self.api_key: raise ValueError("OpenRouter API Key not found.") self.scorer_api_url = DEFAULT_API_URL # For fetching files/submitting self.llm_api_url = HF_API_URL self.model_name = MODEL_NAME # Load system prompt try: with open("prompt.txt", "r") as file: self.system_prompt = file.read().strip() except FileNotFoundError: print("Error: prompt.txt not found. Using a default system prompt.") self.system_prompt = "You are a helpful AI assistant. Please answer the user's questions. Use tools if necessary by outputting TOOL: {\"name\": \"tool_name\", \"args\": {\"arg_name\": \"value\"}}. When you have the final answer, output ANSWER: your_final_answer." except Exception as e: print(f"Error loading prompt.txt: {e}. Using a default system prompt.") self.system_prompt = "You are a helpful AI assistant. Please answer the user's questions. Use tools if necessary by outputting TOOL: {\"name\": \"tool_name\", \"args\": {\"arg_name\": \"value\"}}. When you have the final answer, output ANSWER: your_final_answer." # Define tools self.tools = { "web_search_tool": web_search_tool, "decimal_approximation_tool": decimal_approximation_tool, "get_files_task_id_tool": get_files_task_id_tool # image_processing_tool removed as requested } print(f"Agent tools initialized: {list(self.tools.keys())}") def _call_llm(self, conversation_history: list) -> str: print(f"Calling LLM. Conversation history length: {len(conversation_history)}") # Hugging Face headers (simpler) headers = { "Authorization": f"Bearer {self.api_key}", # should be your HF token "Content-Type": "application/json" } # Format conversation into a single prompt string # You can improve this formatting later if needed prompt = "" for turn in conversation_history: role = turn.get("role", "user").capitalize() content = turn.get("content", "") prompt += f"{role}: {content}\n" prompt += "Assistant:" # Payload for Hugging Face payload = { "inputs": prompt, "parameters": { "temperature": 0.7, "max_new_tokens": 512 } } url = "https://api-inference.huggingface.co/models/deepseek-ai/DeepSeek-V3" response = requests.post(url, headers=headers, json=payload) try: return response.json()[0]["generated_text"].split("Assistant:")[-1].strip() except Exception: return f"Error: {response.text}" def __call__(self, question_data: dict) -> str: task_id = question_data.get("task_id") question_text = question_data.get("question") print(f"Agent received task_id: {task_id}, question (first 50 chars): {str(question_text)[:50]}...") if not task_id or question_text is None: print("Error: Missing task_id or question in agent input.") return "Error: Invalid input to agent." conversation = [ {"role": "system", "content": self.system_prompt}, {"role": "user", "content": question_text} ] max_loops = 1000 # Prevent infinite loops for loop_count in range(max_loops): print(f"\nAgent Loop: {loop_count + 1}") llm_response = self._call_llm(conversation) if llm_response.startswith("ANSWER:"): answer = llm_response[len("ANSWER:"):].strip() print(f"Agent returning final answer: {answer}") return answer elif llm_response.startswith("TOOL:"): tool_call_str = llm_response[len("TOOL:"):].strip() print(f"Attempting tool call: {tool_call_str}") try: tool_data = json.loads(tool_call_str) # Parse the JSON string tool_name = tool_data.get("name") tool_args_dict = tool_data.get("args", {}) if tool_name in self.tools: print(f"Executing tool: {tool_name} with args: {tool_args_dict}") # Special handling for get_files_task_id_tool if it doesn't take generic args if tool_name == "get_files_task_id_tool": # Ensure task_id is passed correctly, not from LLM args unless intended observation = self.tools[tool_name](task_id) else: observation = self.tools[tool_name](**tool_args_dict) print(f"Tool observation: {str(observation)[:200]}...") conversation.append({"role": "assistant", "content": llm_response}) # LLM's tool request conversation.append({"role": "user", "content": f"Observation: {observation}"}) # Tool result else: print(f"Error: Unknown tool name: {tool_name}") conversation.append({"role": "user", "content": f"Error: Unknown tool '{tool_name}'. Available tools are: {', '.join(self.tools.keys())}."}) except json.JSONDecodeError as e: print(f"Error decoding JSON for tool call: {e} - String was: {tool_call_str}") conversation.append({"role": "user", "content": f"Error: Invalid tool call format. Expected JSON. {e}"}) except Exception as e: print(f"Error executing tool or processing its call: {e}") conversation.append({"role": "assistant", "content": llm_response}) # LLM's tool request conversation.append({"role": "user", "content": f"Error executing tool {tool_name}: {e}"}) else: # If the LLM doesn't use the specified prefixes, treat its response as a potential direct answer or a misstep. # Could also be a clarification question from the LLM. print(f"LLM response did not start with ANSWER: or TOOL:. Treating as intermediate thought or error. Response: {llm_response[:100]}") # Adding it as an assistant message and prompting for a structured response conversation.append({"role": "assistant", "content": llm_response}) conversation.append({"role": "user", "content": "Please respond with either 'TOOL: {\"name\": \"tool_name\", \"args\": {}}' or 'ANSWER: your_final_answer'."}) if loop_count == max_loops - 1: print("Agent reached max loops. Returning last LLM response or error.") return f"Error: Agent reached maximum iteration limit. Last response: {llm_response}" return "Error: Agent loop completed without returning an answer." # --- Tool Definitions --- def web_search_tool(search_terms: str) -> str: """ Retrieves information from the internet using DuckDuckGo Search and returns results in JSON format. Args: search_terms (str): The search query to look up. Returns: str: JSON string containing search results. """ print(f"Web search tool called with terms: {search_terms}") try: with DDGS() as ddgs: results = [r for r in ddgs.text(search_terms, max_results=3)] return json.dumps({"results": results}) # Ensure it's a JSON string except Exception as e: print(f"Web search failed: {e}") return json.dumps({"error": f"Search failed: {str(e)}"}) def decimal_approximation_tool(number: float, decimals: int = 1) -> float: """ Adjusts a numerical answer to the specified number of decimal places. Args: number (float): The number to round. decimals (int): Number of decimal places to round to (default is 1). Returns: float: The rounded number. """ print(f"Decimal approximation tool called with number: {number}, decimals: {decimals}") try: return round(float(number), int(decimals)) except Exception as e: print(f"Decimal approximation failed: {e}") return f"Error in decimal_approximation_tool: {str(e)}" # Return error as string def get_files_task_id_tool(task_id: str) -> str: """ Downloads the file associated with the given task_id by making an API call. Args: task_id (str): The ID of the task to fetch the file for. Returns: str: The file content as a string or an error message. """ print(f"Get files tool called with task_id: {task_id}") try: # Assuming DEFAULT_API_URL is the base for the /files endpoint file_url = f"{DEFAULT_API_URL}/files/{task_id}" response = requests.get(file_url, timeout=30) if response.status_code == 200: return response.text else: return f"Error fetching file for task_id {task_id}: Status {response.status_code}, Response: {response.text}" except Exception as e: print(f"Error in get_files_task_id_tool: {e}") return f"Error fetching file for task_id {task_id}: {str(e)}" # --- Gradio App --- def run_and_submit_all(profile: gr.OAuthProfile | None): """ Fetches all questions, runs the BasicAgent on them, submits all answers, and displays the results. """ space_id = os.getenv("SPACE_ID") if profile: username = f"{profile.username}" print(f"User logged in: {username}") else: print("User not logged in.") return "Please Login to Hugging Face with the button.", None scorer_api_url = DEFAULT_API_URL questions_url = f"{scorer_api_url}/questions" submit_url = f"{scorer_api_url}/submit" try: agent = BasicAgent() except Exception as e: print(f"Error instantiating agent: {e}") return f"Error initializing agent: {e}", None agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main" if space_id else "local_run_code_link_not_available" print(f"Agent code link: {agent_code}") print(f"Fetching questions from: {questions_url}") try: response = requests.get(questions_url, timeout=15) response.raise_for_status() questions_data = response.json() if not questions_data: print("Fetched questions list is empty.") return "Fetched questions list is empty or invalid format.", None print(f"Fetched {len(questions_data)} questions.") except requests.exceptions.RequestException as e: print(f"Error fetching questions: {e}") return f"Error fetching questions: {e}", None except requests.exceptions.JSONDecodeError as e: print(f"Error decoding JSON response from questions endpoint: {e}. Response text: {response.text[:500]}") return f"Error decoding server response for questions: {e}", None except Exception as e: print(f"An unexpected error occurred fetching questions: {e}") return f"An unexpected error occurred fetching questions: {e}", None results_log = [] answers_payload = [] print(f"Running agent on {len(questions_data)} questions...") for item in questions_data: task_id = item.get("task_id") question_text = item.get("question") if not task_id or question_text is None: print(f"Skipping item with missing task_id or question: {item}") continue try: # Pass the whole item dictionary to the agent submitted_answer = agent(item) answers_payload.append({"task_id": task_id, "submitted_answer": str(submitted_answer)}) # Ensure answer is string results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": str(submitted_answer)}) except Exception as e: print(f"Error running agent on task {task_id}: {e}") import traceback traceback.print_exc() # Print full traceback for agent errors results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"}) if not answers_payload: print("Agent did not produce any answers to submit.") return "Agent did not produce any answers to submit.", pd.DataFrame(results_log) submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload} status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..." print(status_update) print(f"Submitting {len(answers_payload)} answers to: {submit_url}") try: response = requests.post(submit_url, json=submission_data, timeout=60) response.raise_for_status() result_data = response.json() final_status = ( f"Submission Successful!\n" f"User: {result_data.get('username')}\n" f"Overall Score: {result_data.get('score', 'N/A')}% " f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n" f"Message: {result_data.get('message', 'No message received.')}" ) print("Submission successful.") except requests.exceptions.HTTPError as e: error_detail = f"Server responded with status {e.response.status_code}." try: error_json = e.response.json() error_detail += f" Detail: {error_json.get('detail', e.response.text)}" except requests.exceptions.JSONDecodeError: error_detail += f" Response: {e.response.text[:500]}" final_status = f"Submission Failed: {error_detail}" print(final_status) except requests.exceptions.Timeout: final_status = "Submission Failed: The request timed out." print(final_status) except requests.exceptions.RequestException as e: final_status = f"Submission Failed: Network error - {e}" print(final_status) except Exception as e: final_status = f"An unexpected error occurred during submission: {e}" print(final_status) results_df = pd.DataFrame(results_log) return final_status, results_df with gr.Blocks() as demo: gr.Markdown("# Basic Agent Evaluation Runner") gr.Markdown( """ **Instructions:** 1. Ensure your `TEST_AGENT_KEY` (OpenRouter API Key) is set in your Hugging Face Space secrets or `.env` file. 2. Modify `prompt.txt` to guide the agent, especially for tool use and answer formatting. Remove references to the old `image_processing_tool`. 3. Log in to your Hugging Face account using the button below. 4. Click 'Run Evaluation & Submit All Answers'. --- **Disclaimers:** Agent execution can take time. This setup is a starting point. """ ) gr.LoginButton() run_button = gr.Button("Run Evaluation & Submit All Answers") status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False) results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True) run_button.click( fn=run_and_submit_all, outputs=[status_output, results_table], api_name="run_evaluation" # Added api_name for programmatic access if needed ) if __name__ == "__main__": print("\n" + "-"*30 + " App Starting " + "-"*30) space_host_startup = os.getenv("SPACE_HOST") space_id_startup = os.getenv("SPACE_ID") if space_host_startup: print(f"✅ SPACE_HOST found: {space_host_startup}") print(f" Runtime URL should be: https://{space_host_startup}.hf.space") else: print("ℹ️ SPACE_HOST environment variable not found (running locally?).") if space_id_startup: print(f"✅ SPACE_ID found: {space_id_startup}") print(f" Repo URL: https://huggingface.co/spaces/{space_id_startup}") print(f" Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main") else: print("ℹ️ SPACE_ID environment variable not found (running locally?). Repo URL cannot be determined.") print("-"*(60 + len(" App Starting ")) + "\n") print("Launching Gradio Interface for Basic Agent Evaluation...") demo.launch(debug=True, share=False)