AgentPilot: Make Your Agents Move Better and Faster - HKUDS/AgentPilot

Se describen cinco perfiles de desarrollo software con inteligencia artificial, usando y creando agentes de código.

Se describen cinco perfiles de desarrollo software con inteligencia artificial, usando y creando agentes de código.

Se describen cinco perfiles de desarrollo software con inteligencia artificial, usando y creando agentes de código.

Hmm… It seems to be happening frequently. github.com/huggingface/smolagents #### Repeatedly getting "Error in code parsing: Your code snippet is invalid" opened 12:41PM - 15 Jan 25 UTC rnckp I repeatedly get this error when using the Codeagent: ```Error in code parsing:… The code blob is invalid, because the regex pattern ```(?:py|python)?\n(.*?)\n``` was not found in code_blob ... ``` I have tried to emphasize proper code block formatting in the agent.run() argument, to no avail. The error occurs while using GPT-4o and Sonnet 3.5. How can I fix this or at least improve realiability? Thanks for any help in this regard!

Hello, I am trying to train a forecast model with a CodeAgent but it fails. It shows `Code execution failed at line 'system` due to : InterpreterError: The variable ‘system’ is not defined` or other times is shows `s` instead of `system`. The code: import pandas as pd import numpy as np import matplotlib.pyplot as plt import json import os import re import traceback from datetime import datetime, timedelta from prophet import Prophet from smolagents import CodeAgent, TransformersModel, LogLevel, tool from transformers import AutoTokenizer # Create output directories os.makedirs('forecast_outputs', exist_ok=True) os.makedirs('forecast_plots', exist_ok=True) # Initialize the Qwen model (same as in your original code) def setup_qwen_model(): """Initialize the Qwen model for agents""" try: model_id = "cmarkea/Qwen2.5-Coder-7B-Instruct-4bit" tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True) # Fix the pad token issue if tokenizer.pad_token_id is None: tokenizer.pad_token_id = tokenizer.eos_token_id model = TransformersModel( model_id=model_id, device_map="auto", torch_dtype="auto", max_new_tokens=512, temperature=0.2, trust_remote_code=True, tokenizer=tokenizer, pad_token_id=tokenizer.pad_token_id, eos_token_id=tokenizer.eos_token_id ) return model except Exception as e: print(f"Error loading model: {str(e)}") print(traceback.format_exc()) raise e # Create a sample weather dataframe def create_sample_weather_data(): # Generate 30 days of hourly data start_date = datetime(2025, 4, 15) periods = 30 * 24 # 30 days of hourly data dates = [start_date + timedelta(hours=i) for i in range(periods)] # Generate synthetic temperature data with daily pattern and some trend temps = [] for i in range(periods): hour = i % 24 day = i // 24 # Base temperature with seasonal and hourly variation base_temp = 20 + 5 * np.sin(np.pi * day / 15) hourly_variation = 5 * np.sin(np.pi * hour / 12) random_factor = np.random.normal(0, 1) temps.append(base_temp + hourly_variation + random_factor) # Create DataFrame df = pd.DataFrame({ 'DateTime': dates, 'Temperature': temps, 'Humidity': [40 + 20 * np.sin(np.pi * i / 24) + np.random.normal(0, 5) for i in range(periods)], 'Pressure': [1013 + 5 * np.sin(np.pi * i / 48) + np.random.normal(0, 1) for i in range(periods)] }) # Save to CSV csv_path = 'forecast_outputs/sample_weather_data.csv' df.to_csv(csv_path, index=False) return csv_path # Define required tools for the agent @tool def create_forecast_model(csv_path: str, date_column: str, target_column: str, periods: int = 30) -> str: """ Create a Prophet forecast model and generate predictions. Args: csv_path: Path to the CSV file date_column: Name of the date column target_column: Name of the target column to forecast periods: Number of periods to forecast Returns: JSON string with forecast results """ try: print(f"Creating forecast for {csv_path}, date: {date_column}, target: {target_column}") df = pd.read_csv(csv_path) # Prepare data for Prophet prophet_df = pd.DataFrame({ 'ds': pd.to_datetime(df[date_column]), 'y': df[target_column] }) # Remove any rows with missing values prophet_df = prophet_df.dropna() # Create and fit the model model = Prophet() model.fit(prophet_df) # Make future dataframe future = model.make_future_dataframe(periods=periods) # Generate forecast forecast = model.predict(future) # Save results forecast_path = 'forecast_outputs/forecast_results.csv' forecast.to_csv(forecast_path, index=False) # Generate plot plt.figure(figsize=(12, 6)) model.plot(forecast) plt.title(f'Prophet Forecast of {target_column}') plt.tight_layout() plot_path = 'forecast_plots/forecast_plot.png' plt.savefig(plot_path) plt.close() # Return summary results = { "success": True, "periods_forecasted": periods, "forecast_path": forecast_path, "plot_path": plot_path, "original_data_points": len(prophet_df) } return json.dumps(results, indent=2) except Exception as e: import traceback error_msg = f"Error creating forecast: {str(e)}" print(error_msg) print(f"Traceback: {traceback.format_exc()}") return json.dumps({"error": error_msg}) # Create a forecasting agent with Qwen model def create_forecasting_agent(model): agent = CodeAgent( model=model, tools=[create_forecast_model], additional_authorized_imports=["pandas", "numpy", "prophet", "matplotlib"], max_steps=3, verbosity_level=LogLevel.INFO, planning_interval=2 ) return agent # Main function def main(): print("Setting up Weather Forecasting Agent Example with Qwen Model") # Create sample data csv_path = create_sample_weather_data() model = setup_qwen_model() # Create agent with the loaded model agent = create_forecasting_agent(model) print("Created forecasting agent with Qwen model") prompt = f""" You are a weather forecasting assistant. I need you to: 1. Create a 7-day forecast for Temperature using the weather data in {csv_path} 2. Use the create_forecast_model tool to generate the forecast 3. Set the date_column to 'DateTime', the target_column to 'Temperature', and periods to 7 Return the results of the forecast, including any statistics and where the output files are saved. """ result = agent.run(prompt) print(result) # Try to parse any JSON in the result try: # Look for JSON content in the result json_pattern = r'\{.*\}' json_match = re.search(json_pattern, result, re.DOTALL) if json_match: forecast_result = json.loads(json_match.group()) if 'success' in forecast_result and forecast_result['success']: print("\nForecast created successfully!") print(f"Forecast saved to: {forecast_result['forecast_path']}") print(f"Plot saved to: {forecast_result['plot_path']}") # Load and display some statistics from the forecast forecast_df = pd.read_csv(forecast_result['forecast_path']) print("\nForecast Statistics:") print(f"Total periods: {len(forecast_df)}") print(f"Forecast start: {forecast_df['ds'].min()}") print(f"Forecast end: {forecast_df['ds'].max()}") print(f"Average predicted value: {forecast_df['yhat'].mean():.2f}") print(f"Minimum predicted value: {forecast_df['yhat'].min():.2f}") print(f"Maximum predicted value: {forecast_df['yhat'].max():.2f}") except Exception as e: print(f"Error extracting forecast details: {str(e)}") print("Please check the agent's output for forecast information") if __name__ == "__main__": main() generally, i can see this error (code execution failed at line) at other cases I have created but sometimes it shwos that sometimes not and sometimes the whole execution goes well even though this error exists. But now, it doesn’t proceed. I am attaching some screenshots.

Perhaps this issue? github.com/huggingface/smolagents #### Multi Agent code execution failed - InterpreterError: It is not permitted to evaluate other functions than the provided tools or functions defined/imported in previous code opened 03:37PM - 13 Feb 25 UTC closed 02:19PM - 24 Feb 25 UTC skown Hey, I'm getting this error all the time: **InterpreterError:It is not permitte…d to evaluate other functions than the provided tools or functions defined/imported in previous code** In manager.run I'm trying to pass additional_args=dict(source_file="file.csv"). Example of agent: ``` agent_core = CodeAgent( model = model, tools = [tool_1, tool_2], additional_authorized_imports=["pandas", "matplotlib", "seaborn"], name="Data Analyst", description="""Analysis agent with a set of ready-to-use tools that parse, cleanse, and process the data provided.""" ) ``` ``` manager_agent = CodeAgent( model=model, tools=[], managed_agents=[agent_core], additional_authorized_imports=["pandas", "matplotlib", "seaborn"], ) ``` It works normally when I try to call agent_core.run directly, without manager_agent. Does anyone have a similar problem?

I am trying to use the Azure Openai’s “GPT 4-o-mini” LLM model, from the “SAP AI Core” service in the SAP Business Technology Platform, when developing an Agentic AI application using SMOL Agents. The problem is models in SAP AI Core service can be called like an API using the “request” library for text generation, and here “bearer” token-based Authentication mechanism is used where the token is passed to the header variable of the post method of “request” library. But in “CodeAgents” initialization or “LiteLLM” based initializations, model name and openai-api-key based models are only passed. Therefore, there is no way to call or load an LLM Model hosted in “SAP AI Core” service. Because of this we are unable to use the SMOL Agents framework for our Agentic AI application. Could anyone please help deal with this issue?