Chapter 2 questions

Course
68

try change the tool description, so it only will take one word as Argument.

Args:
occasion: One word representung the type of occasion for the party menu.

1 Like
69
70

Hello!
does anyone know, or can point me to any documentation on the following regard?
I see in the examples from HF course Lesson 1 and lesson 2, specifically in the Alfred agent, (sergiopaniego/AlfredAgent at main) that there are the agent.json and the promts.yaml files, and I canā€™t understand why they are for and how they work. I did multiple tests, and I see the same behaviour in my agents using or not using those two files. Any help will be appreciated.
Also, a part from knowing what they are for and how they effect the agent, is there any template that should be followed to create those files in terms of the structure or allowed keys?
Thank you.

71

Iā€™ll answer myself.
So the prompts.yaml works as expected, but there is no changes when testing it with or without it, because the CodeAgent, automatically loads the same prompt template from its own resource in the smolagents library when there is no value passed in the prompt_templates argument.
Now, for the agent.json seems to be from a functionality that exports your agent to a json file, called basically when you push to HF repo (push_to_hub which already creates the prompts.yaml as well) so seems a way to be able to re-load it later on from a folder or from the hub itself.

1 Like
72

In the langfuse logging of the agent of 2.1 ā€™ Building Agents That Use Codeā€™ you can see that there is an import error of the ā€˜reā€™ module. I have already created an issue on github, but wanted to document it here too for anybody wondering.

Amazing though how the agent handles the error and still responds with a valid answer.

1 Like
73

hello everybody, happy new year. I am an impostor coder trying to learn.

2 Likes
74

In smolagents ā€œBuilding Agents That Use Codeā€ cell 12, Iā€™m getting the following error - would appreciate help please:

from smolagents import CodeAgent, HfApiModel, Tool

image_generation_tool = Tool.from_space(
ā€œblack-forest-labs/FLUX.1-schnellā€,
name=ā€œimage_generatorā€,
description=ā€œGenerate an image from a promptā€
)

model = HfApiModel(ā€œQwen/Qwen2.5-Coder-32B-Instructā€)

agent = CodeAgent(tools=[image_generation_tool], model=model)

agent.run(
ā€œImprove this prompt, then generate an image of it.ā€,
additional_args={ā€˜user_promptā€™: ā€˜A grand superhero-themed party at Wayne Manor, with Alfred overseeing a luxurious galaā€™}
)

1 Like
75

I am a bit lost, on the vision agents we are using gpt-4oā€¦ but how do I get a free API key? Or the idea is not to run the example?

76

404 page for notebook link from here

1 Like
77

have you tried this? Chapter 2 questions - #66 by gael1130

78

How do I get a Free Api key to do the example?

79

How do I get a Free gpt 4o Api key to do the example?

80

ok, here Iā€™ve made mine

81

Issue with CodeAgent Initialization in Multi-Agent System (Chapter 2) of Hugging Face Agents Course

Hi, I am encountering multiple errors while working through Chapter 2 of the Agents Course on Hugging Face. Specifically, Iā€™m facing issues in the section Splitting the Task Between Two Agents.

I am using LiteLLMModel as shown below:

from smolagents import LiteLLMModel
from dotenv import load_dotenv

load_dotenv()

messages = []
model = LiteLLMModel("gemini/gemini-2.0-flash-exp", temperature=0.2)

When I run the following code block(the tool: calculate_cargo_travel_time is already defined in earlier section):

from smolagents import (
    CodeAgent,
    DuckDuckGoSearchTool,
    VisitWebpageTool
)

web_agent = CodeAgent(
    model=model,
    tools=[
        DuckDuckGoSearchTool(),
        VisitWebpageTool(),
        calculate_cargo_travel_time,
    ],
    name="web_agent",
    description="Browses the web to find information",
    verbosity_level=0,
    max_steps=10,
)

I get the following error:

TypeError: MultiStepAgent.__init__() got an unexpected keyword argument 'final_answer_checks'

Additionally, I encountered a similar issue when commenting out the name parameter, which results in the same error for the description parameter.

Also, in another section of the code:

from smolagents.utils import encode_image_base64, make_image_url
from smolagents import OpenAIServerModel

def check_reasoning_and_plot(final_answer, agent_memory):
    # Function definition here
    output = multimodal_model(messages).content
    print("Feedback: ", output)
    if "FAIL" in output:
        raise Exception(output)
    return True

manager_agent = CodeAgent(
    model=model,
    tools=[calculate_cargo_travel_time],
    managed_agents=[web_agent],
    additional_authorized_imports=[
        "geopandas",
        "plotly",
        "shapely",
        "json",
        "pandas",
        "numpy",
    ],
    planning_interval=5,
    verbosity_level=2,
    final_answer_checks=[check_reasoning_and_plot],
    max_steps=15,
)

I receive the error:

TypeError: MultiStepAgent.__init__() got an unexpected keyword argument 'final_answer_checks'

Could anyone help me resolve these issues or provide guidance on how to fix them? Thank you in advance!

1 Like
82

Possibly version issue?

83

I am getting this error while running the code in colag.
Error in generating model output:
402 Client Error: Payment Required for url:
https://api-inference.huggingface.co/models/Qwen/Qwen2.5-Coder-32B-Instruct/v1/chat/completions (Request ID:
Root=1-67d1d83b-444897b474d9f00a6aca3e66;f5647a7f-2be3-4b1c-8f87-c825587e4044)

You have exceeded your monthly included credits for Inference Providers. Subscribe to PRO to get 20x more monthly
included credits.
[Step 1: Duration 0.23 seconds

3 Likes
84

There are two possibilities for a 402 error: either there is a problem on the HF side, or it is a normal result that simply exceeds the available amount.

85

I am getting the same error as well. Is there a requirement to subscribe to PRO? Would like to understand the cost implications of such a subscription.

1 Like
86

Itā€™s because I have hit Hugging Faceā€™s free quota limit for the Qwen2.5-Coder-32B-Instruct model. Either I have to subscribe to their PRO plan for more credits, or run the model locally.

1 Like
87

I am trying to run AlfredAgent its giving me error Step 1

:collision: Error

Error in generating model output:
422 Client Error: Unprocessable Entity for url: https://router.huggingface.co/hf-inference/models/Qwen/Qwen2.5-Coder-32B-Instruct/v1/chat/completions (Request ID: Root=1-67d497ef-6276d83d2d38b86f1ab93d5a;2925a3a7-2d62-421a-b690-3f88caa0d4b5)

Input validation error: inputs tokens + max_new_tokens must be <= 32768. Given: 250290 inputs tokens and 0 max_new_tokens
{ā€œerrorā€:ā€œInput validation error: inputs tokens + max_new_tokens must be <= 32768. Given: 250290 inputs tokens and 0 max_new_tokensā€,ā€œerror_typeā€:ā€œvalidationā€}

1 Like