crewAI: "Action don't exist" when using Ollama.

I am running the example script from the readme. Other than my API keys, the only change I made was to have the script run off my local LLM.

When I run it, I get a Action 'Search the internet using the query "Latest Advancements in AI 2024".' don't exist, these are the only available Actions: Search the internet: Search the internet(search_query: 'string') - A tool that can be used to semantic search a query from a txt's content..

Any assistance would be appreciated.

Code below.

import os
from langchain_community.llms import Ollama
from crewai import Agent, Task, Crew, Process
from crewai_tools import SerperDevTool

os.environ["OPENAI_API_KEY"] = "<my_openai_api_key>"
os.environ["SERPER_API_KEY"] = "<my_serper_api_key>" # serper.dev API key

# You can choose to use a local model through Ollama for example. See https://docs.crewai.com/how-to/LLM-Connections/ for more information.

os.environ["OPENAI_API_BASE"] = 'http://localhost:11434/v1'
os.environ["OPENAI_MODEL_NAME"] ='llama2'  # Adjust based on available model
os.environ["OPENAI_API_KEY"] =''
llama2 = Ollama(model="llama2")

search_tool = SerperDevTool()

# Define your agents with roles and goals
researcher = Agent(
  role='Senior Research Analyst',
  goal='Uncover cutting-edge developments in AI and data science',
  backstory="""You work at a leading tech think tank.
  Your expertise lies in identifying emerging trends.
  You have a knack for dissecting complex data and presenting actionable insights.""",
  verbose=True,
  allow_delegation=False,
  tools=[search_tool],
  # You can pass an optional llm attribute specifying what mode you wanna use.
  # It can be a local model through Ollama / LM Studio or a remote
  # model like OpenAI, Mistral, Antrophic or others (https://docs.crewai.com/how-to/LLM-Connections/)
  #
  # import os
  # os.environ['OPENAI_MODEL_NAME'] = 'gpt-3.5-turbo'
  #
  # OR
  #
  # from langchain_openai import ChatOpenAI
  # llm=ChatOpenAI(model_name="gpt-3.5", temperature=0.7)
  llm=llama2
)
writer = Agent(
  role='Tech Content Strategist',
  goal='Craft compelling content on tech advancements',
  backstory="""You are a renowned Content Strategist, known for your insightful and engaging articles.
  You transform complex concepts into compelling narratives.""",
  verbose=True,
  allow_delegation=True,
  llm=llama2
)

# Create tasks for your agents
task1 = Task(
  description="""Conduct a comprehensive analysis of the latest advancements in AI in 2024.
  Identify key trends, breakthrough technologies, and potential industry impacts.""",
  expected_output="Full analysis report in bullet points",
  agent=researcher
)

task2 = Task(
  description="""Using the insights provided, develop an engaging blog
  post that highlights the most significant AI advancements.
  Your post should be informative yet accessible, catering to a tech-savvy audience.
  Make it sound cool, avoid complex words so it doesn't sound like AI.""",
  expected_output="Full blog post of at least 4 paragraphs",
  agent=writer
)

# Instantiate your crew with a sequential process
crew = Crew(
  agents=[researcher, writer],
  tasks=[task1, task2],
  verbose=2, # You can set it to 1 or 2 to different logging levels
)

# Get your crew to work!
result = crew.kickoff()

print("######################")
print(result)

Terminal output (I ctrl + c after the action message)

(.venv) me@my-mbp-2 crewai-testing % python crewai_test.py 
 [DEBUG]: == Working Agent: Senior Research Analyst
 [INFO]: == Starting Task: Conduct a comprehensive analysis of the latest advancements in AI in 2024.
  Identify key trends, breakthrough technologies, and potential industry impacts.


> Entering new CrewAgentExecutor chain...
Thought: I understand the task at hand, which is to conduct a comprehensive analysis of the latest advancements in AI in 2024. I will use the internet search tool to gather information and present my findings in a bullet point format.

Action: Search the internet using the query "Latest Advancements in AI 2024".
Action Input: {
"search_query": "Latest Advancements in AI 2024",
"tool": "Search the internet"
} 

Action 'Search the internet using the query "Latest Advancements in AI 2024".' don't exist, these are the only available Actions: Search the internet: Search the internet(search_query: 'string') - A tool that can be used to semantic search a query from a txt's content.

About this issue

  • Original URL
  • State: closed
  • Created 3 years ago
  • Reactions: 11
  • Comments: 17 (1 by maintainers)

Most upvoted comments

This did work at one point, but now it’s broken. This is very disappointing.

I get it, maybe it’s frustrating but this is open source, comments like this one must have that in mind, and maybe a PR that solves it… Or at least something to debug. That would be a constructive, significant contribution to a free, unpaid job that João makes here.

This did work at one point, but now it’s broken. This is very disappointing.

now Im focusing on task delegation only, without tools. But like @kyuumeitai said, something that is really producing less errors during execution is using ‘temperature’ values close to zero.

This “seems” to make the LLM follow more accurately the instructions inside crewai code. My only concern is that may limit the “creativity” of the model.

Right. That is basically a function of temperature. You might want to configure “creative” agents to use LLMs that don’t have low temperature. Not all agents need to use the same LLM.

Honestly, I think there are a few pieces to the puzzle here: (I’m still learning)

  1. The LLM of choice should be able to understand the concept of “function calling”. Not all of them do. I can use mistral just fine but I think I need to turn down the temperature to 0.5 or less for it to get more consistent.
  2. The API wrapper of choice should support “tools”. I have seen the best results with ChatOpenAI from langchain_openai library. I have seen OllamaFunctions from langchain_experimental.llms as well but I just run with ChatOpenAI so i can swap between Gpt-4, gpt-3.5, and my local Ollama / Mistral.

I gained a good bit of clarity after going through this fella’s tutorials on Langchain proper, but I’m still pretty new. ( https://www.youtube.com/playlist?list=PLcQVY5V2UY4Kat6vxC7ESzIIzHWdwlnak )