CLI crewai train throws error 'The filename must not end with .pkl' even with .py file

General
1

Hey folks, how’s it going? I’m trying to use the CLI crewai train to train my medical agent, but I keep hitting a weird error I can’t figure out. The command I’m running is: crewai train -f medical_agent_train.py -n 3. But every time, I get this: Training the Crew for 3 iterations - An unexpected error occurred: The filename must not end with .pkl. The file medical_agent_train.py is a valid Python script, not a .pkl, so I don’t get why the CLI is rejecting it. I’ve tried a bunch of variations: crewai train -f medical_agent_train -n 3, crewai train -f “C:\Users\jaian\Desktop\cla\medical_agent_train.py” -n 3, renamed it to train_medico.py and ran crewai train -f train_medico -n 3, even tried python -m crewai.cli train -f C:\Users\jaian\Desktop\cla\medical_agent_train.py -n 3. Still, same error every time. The --help shows -f and -n are valid options: Usage: crewai train [OPTIONS] Options: -n, --n_iterations INTEGER Number of iterations to train the crew -f, --filename TEXT Path to a custom file for training --help Show this message and exit. Looks like a bug in the CLI to me, since the file is .py, not .pkl. The docs (Introduction - CrewAI) say crewai train -n <n_iterations> should work with a Python script, but it’s not happening. Anyone run into this or know how to fix it?

2

Try:

crewai train -n 3 -f train_medico.pkl

This work on my machine with v0.108.0

3

Training the Crew for 3 iterations
error: Failed to spawn: train
Caused by: program not found
An error occurred while training the crew: Command ‘[‘uv’, ‘run’, ‘train’, ‘3’, ‘train_medico.pkl’]’ returned non-zero exit status 2.

4

What version of CrewAI are you using and can you paste the contents of your pyproject.toml file?

Are you in a CrewAI CLI created crew?

5

0.108.0

import os
from dotenv import load_dotenv
from crewai import Agent, Task, Crew, Process

Load the .env file

load_dotenv()
print(“OpenAI API Key loaded:”, os.getenv(“OPENAI_API_KEY”))

Define the medical agent

medical_agent = Agent(
role=“Medical Specialist”,
goal=“Assist patients with medical queries and provide accurate guidance”,
backstory=“You are an experienced doctor with years of practice, specialized in diagnosis and treatment. Use a professional, empathetic, and clear tone.”,
verbose=True,
llm=“gpt-4o”,
memory=False,
allow_delegation=False
)

Define the task for training

task = Task(
description=“Respond to the patient’s query: ‘I have mild chest pain’. Provide an accurate and helpful medical response.”,
agent=medical_agent,
expected_output=“A clear and professional response to the patient”
)

Class to manage the Crew

class MedicalCrew:
def init(self):
self.agents = [medical_agent]
self.tasks = [task]

@crew
def crew(self) -> Crew:
    """Creates the Medical Crew for training"""
    crew_instance = Crew(
        agents=self.agents,
        tasks=self.tasks,
        process=Process.sequential,
        verbose=True,
        memory=False  # No Mem0, just behavior training
    )
    # Try to load trained data, if it exists
    output = self.agents[0]._use_trained_data(".")
    print("Trained data:", output)
    return crew_instance

Function to interact in the terminal after training

def interact_with_agent(crew):
print(“Welcome! I’m your trained medical specialist. How can I assist you today?”)
while True:
user_input = input(“You: “)
if user_input.lower() in [“exit”, “quit”, “bye”]:
print(“Goodbye! Take care!”)
break
task.description = f"Respond to the patient’s query: ‘{user_input}’. Provide an accurate and helpful medical response.”
response = crew.kickoff()
print(f"Doctor: {response}”)

Execute

if name == “main”:
# Create the Crew
medical_crew = MedicalCrew()
crew_instance = medical_crew.crew()
# Run the interaction
interact_with_agent(crew_instance)

6

Did you create the project via crewai create crew your_crew?

7

Yes

image
image190×513 8.99 KB