Smolagents a Smol Library to build great Agents

Are you waiting for the future or creating it?

Be a part of Gen AI Launch Pad 2024 and take charge of what’s next. Act today for a better tomorrow.

Introduction

smolagents is a lightweight library for constructing and running agents with just a few lines of code. It emphasizes simplicity, security, and versatility by providing:

Code Agents: Agents that execute actions through code.
ToolCalling Agents: Agents that output actions as JSON or text.
Integration with multiple language models (LLMs), including OpenAI, Anthropic, and Hugging Face.

In this blog, we will:

Explore the key features of smolagents.
Analyze various examples, including building a search tool, creating custom tools, and running a Gradio interface.
Offer real-world applications and resources to deepen your knowledge.

Detailed Explanation

1. Setting Up smolagents

Before diving into examples, we start by installing the required libraries and configuring API keys. Here's the setup process:

Code Snippet

!pip install smolagents datasets langchain langchain_community rank_bm25

import os
from google.colab import userdata

os.environ["OPENROUTER_API_KEY"] = userdata.get('OPENROUTER_API_KEY')
os.environ["ANTHROPIC_API_KEY"] = userdata.get('ANTHROPIC_API_KEY')
os.environ['HF_TOKEN'] = userdata.get('HF_TOKEN')

Explanation

Install required dependencies for smolagents and its integrations.
Set up environment variables to manage API keys securely.

Expected Output

No visual output is expected from this block. Ensure dependencies are installed and keys are correctly configured.

Real-World Application

This setup is critical for leveraging external APIs and tools in smolagents, enabling secure and seamless communication with LLMs.

2. Basic Usage of Code Agents

A CodeAgent executes Python code as part of its reasoning process. Below is an example using a DuckDuckGo search tool and a Hugging Face model.

Code Snippet

from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiModel

agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=HfApiModel())

agent.run("What is the cube root of 27?")

Explanation

CodeAgent: The main agent class for executing Python code.
DuckDuckGoSearchTool: A tool for performing web searches.
HfApiModel: A wrapper for Hugging Face models.

Expected Output

Out - Final answer: 3.0

The agent calculates the cube root of 27 using Python code and provides the correct answer.

Real-World Application

CodeAgents are ideal for automating tasks that require computation or integration with external tools, such as data analysis or web scraping.

3. Custom Tools

You can extend smolagents by creating custom tools. Below is an example of a tool to fetch the most downloaded model for a given task on the Hugging Face Hub.

Code Snippet

from smolagents import Tool

class HFModelDownloadsTool(Tool):
    name = "model_download_counter"
    description = """
    This tool returns the most downloaded model of a given task on the Hugging Face Hub.
    It returns the name of the checkpoint."""
    inputs = {"task": {"type": "string", "description": "The task category."}}
    output_type = "string"

    def forward(self, task: str):
        from huggingface_hub import list_models
        model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
        return model.id

model_downloads_tool = HFModelDownloadsTool()

model = HfApiModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct")

agent = CodeAgent(tools=[HFModelDownloadsTool()], model=model)
agent.run("What is the most downloaded model for text classification?")

Explanation

HFModelDownloadsTool: Fetches the most popular model for a given task.
list_models: A Hugging Face function to search and sort models.

Expected Output

The name of the most downloaded model for the specified task.

Real-World Application

Custom tools allow you to extend smolagents for domain-specific tasks, such as retrieving information from APIs or databases.

4. Gradio User Interface

A simple Gradio interface can be created for the agent, enabling an interactive user experience.

Code Snippet

from smolagents import GradioUI

ui = GradioUI(agent)
ui.launch()

Explanation

GradioUI: Provides an easy-to-use graphical interface for interacting with agents.
ui.launch(): Launches the interface in a web browser.

Expected Output

A Gradio interface with an input box for user queries and an output area for agent responses.

Real-World Application

Use Gradio to create accessible demos or deploy interactive tools for end-users.

5. Retrieval-Augmented Generation (RAG)

Combine retrieval techniques with language models for information-heavy tasks. Below is an example using a custom retriever tool.

Code Snippet

import datasets
from langchain.docstore.document import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.retrievers import BM25Retriever

knowledge_base = datasets.load_dataset("m-ric/huggingface_doc", split="train")
knowledge_base = knowledge_base.filter(lambda row: row["source"].startswith("huggingface/transformers"))

source_docs = [
    Document(page_content=doc["text"], metadata={"source": doc["source"].split("/")[1]})
    for doc in knowledge_base
]

text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
docs_processed = text_splitter.split_documents(source_docs)

class RetrieverTool(Tool):
    name = "retriever"
    description = "Retrieves relevant parts of the Transformers documentation."
    inputs = {"query": {"type": "string", "description": "Search query."}}
    output_type = "string"

    def __init__(self, docs, **kwargs):
        super().__init__(**kwargs)
        self.retriever = BM25Retriever.from_documents(docs, k=10)

    def forward(self, query: str) -> str:
        docs = self.retriever.invoke(query)
        return "\n".join([f"Document {i}: {doc.page_content}" for i, doc in enumerate(docs)])

retriever_tool = RetrieverTool(docs_processed)
agent = CodeAgent(tools=[retriever_tool], model=model)
agent_output = agent.run("Which is slower in Transformers training, the forward or backward pass?")
print(agent_output)

Explanation

BM25Retriever: Performs semantic search on a preprocessed knowledge base.
RecursiveCharacterTextSplitter: Splits documents into manageable chunks.

Expected Output

Relevant documents from the Transformers documentation, highlighting differences between the forward and backward passes.

Real-World Application

RAG is useful for tasks requiring precise answers from large knowledge bases, such as technical documentation or academic research.

Conclusion

The smolagents library simplifies the creation of intelligent agents by combining modularity with the power of LLMs. From basic computation to complex retrieval tasks, smolagents empowers developers to integrate AI into various applications effortlessly.

Key Takeaways

Smolagents makes building agents intuitive and efficient.
It supports various LLMs and custom tools for diverse use cases.
Features like Gradio integration and RAG enable real-world applications across industries.

Next Steps

Explore the smolagents GitHub repository to dive deeper into its capabilities.
Experiment with advanced use cases, such as multi-agent collaboration or API integrations.

Resources

---------------------------------

Stay Updated:- Follow Build Fast with AI pages for all the latest AI updates and resources.

Experts predict 2025 will be the defining year for Gen AI implementation.Want to be ahead of the curve?

Join Build Fast with AI’s Gen AI Launch Pad 2025 - your accelerated path to mastering AI tools and building revolutionary applications.

Are you waiting for the future or creating it?

Be a part of Gen AI Launch Pad 2024 and take charge of what’s next. Act today for a better tomorrow.

Introduction

smolagents is a lightweight library for constructing and running agents with just a few lines of code. It emphasizes simplicity, security, and versatility by providing:

Code Agents: Agents that execute actions through code.
ToolCalling Agents: Agents that output actions as JSON or text.
Integration with multiple language models (LLMs), including OpenAI, Anthropic, and Hugging Face.

In this blog, we will:

Explore the key features of smolagents.
Analyze various examples, including building a search tool, creating custom tools, and running a Gradio interface.
Offer real-world applications and resources to deepen your knowledge.

Detailed Explanation

1. Setting Up smolagents

Before diving into examples, we start by installing the required libraries and configuring API keys. Here's the setup process:

Code Snippet

!pip install smolagents datasets langchain langchain_community rank_bm25

import os
from google.colab import userdata

os.environ["OPENROUTER_API_KEY"] = userdata.get('OPENROUTER_API_KEY')
os.environ["ANTHROPIC_API_KEY"] = userdata.get('ANTHROPIC_API_KEY')
os.environ['HF_TOKEN'] = userdata.get('HF_TOKEN')

Explanation

Install required dependencies for smolagents and its integrations.
Set up environment variables to manage API keys securely.

Expected Output

No visual output is expected from this block. Ensure dependencies are installed and keys are correctly configured.

Real-World Application

This setup is critical for leveraging external APIs and tools in smolagents, enabling secure and seamless communication with LLMs.

2. Basic Usage of Code Agents

A CodeAgent executes Python code as part of its reasoning process. Below is an example using a DuckDuckGo search tool and a Hugging Face model.

Code Snippet

from smolagents import CodeAgent, DuckDuckGoSearchTool, HfApiModel

agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=HfApiModel())

agent.run("What is the cube root of 27?")

Explanation

CodeAgent: The main agent class for executing Python code.
DuckDuckGoSearchTool: A tool for performing web searches.
HfApiModel: A wrapper for Hugging Face models.

Expected Output

Out - Final answer: 3.0

The agent calculates the cube root of 27 using Python code and provides the correct answer.

Real-World Application

CodeAgents are ideal for automating tasks that require computation or integration with external tools, such as data analysis or web scraping.

3. Custom Tools

You can extend smolagents by creating custom tools. Below is an example of a tool to fetch the most downloaded model for a given task on the Hugging Face Hub.

Code Snippet

from smolagents import Tool

class HFModelDownloadsTool(Tool):
    name = "model_download_counter"
    description = """
    This tool returns the most downloaded model of a given task on the Hugging Face Hub.
    It returns the name of the checkpoint."""
    inputs = {"task": {"type": "string", "description": "The task category."}}
    output_type = "string"

    def forward(self, task: str):
        from huggingface_hub import list_models
        model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
        return model.id

model_downloads_tool = HFModelDownloadsTool()

model = HfApiModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct")

agent = CodeAgent(tools=[HFModelDownloadsTool()], model=model)
agent.run("What is the most downloaded model for text classification?")

Explanation

HFModelDownloadsTool: Fetches the most popular model for a given task.
list_models: A Hugging Face function to search and sort models.

Expected Output

The name of the most downloaded model for the specified task.

Real-World Application

Custom tools allow you to extend smolagents for domain-specific tasks, such as retrieving information from APIs or databases.

4. Gradio User Interface

A simple Gradio interface can be created for the agent, enabling an interactive user experience.

Code Snippet

from smolagents import GradioUI

ui = GradioUI(agent)
ui.launch()

Explanation

GradioUI: Provides an easy-to-use graphical interface for interacting with agents.
ui.launch(): Launches the interface in a web browser.

Expected Output

A Gradio interface with an input box for user queries and an output area for agent responses.

Real-World Application

Use Gradio to create accessible demos or deploy interactive tools for end-users.

5. Retrieval-Augmented Generation (RAG)

Combine retrieval techniques with language models for information-heavy tasks. Below is an example using a custom retriever tool.

Code Snippet

import datasets
from langchain.docstore.document import Document
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.retrievers import BM25Retriever

knowledge_base = datasets.load_dataset("m-ric/huggingface_doc", split="train")
knowledge_base = knowledge_base.filter(lambda row: row["source"].startswith("huggingface/transformers"))

source_docs = [
    Document(page_content=doc["text"], metadata={"source": doc["source"].split("/")[1]})
    for doc in knowledge_base
]

text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50)
docs_processed = text_splitter.split_documents(source_docs)

class RetrieverTool(Tool):
    name = "retriever"
    description = "Retrieves relevant parts of the Transformers documentation."
    inputs = {"query": {"type": "string", "description": "Search query."}}
    output_type = "string"

    def __init__(self, docs, **kwargs):
        super().__init__(**kwargs)
        self.retriever = BM25Retriever.from_documents(docs, k=10)

    def forward(self, query: str) -> str:
        docs = self.retriever.invoke(query)
        return "\n".join([f"Document {i}: {doc.page_content}" for i, doc in enumerate(docs)])

retriever_tool = RetrieverTool(docs_processed)
agent = CodeAgent(tools=[retriever_tool], model=model)
agent_output = agent.run("Which is slower in Transformers training, the forward or backward pass?")
print(agent_output)

Explanation

BM25Retriever: Performs semantic search on a preprocessed knowledge base.
RecursiveCharacterTextSplitter: Splits documents into manageable chunks.

Expected Output

Relevant documents from the Transformers documentation, highlighting differences between the forward and backward passes.

Real-World Application

RAG is useful for tasks requiring precise answers from large knowledge bases, such as technical documentation or academic research.

Conclusion

Key Takeaways

Smolagents makes building agents intuitive and efficient.
It supports various LLMs and custom tools for diverse use cases.
Features like Gradio integration and RAG enable real-world applications across industries.

Next Steps

Explore the smolagents GitHub repository to dive deeper into its capabilities.
Experiment with advanced use cases, such as multi-agent collaboration or API integrations.

Resources

---------------------------------

Stay Updated:- Follow Build Fast with AI pages for all the latest AI updates and resources.

Experts predict 2025 will be the defining year for Gen AI implementation.Want to be ahead of the curve?

Join Build Fast with AI’s Gen AI Launch Pad 2025 - your accelerated path to mastering AI tools and building revolutionary applications.

BuildFast Bot

Introduction

Detailed Explanation

1. Setting Up smolagents

Code Snippet

Explanation

Expected Output

Real-World Application

2. Basic Usage of Code Agents

Code Snippet

Explanation

Expected Output

Real-World Application

3. Custom Tools

Code Snippet

Explanation

Expected Output

Real-World Application

4. Gradio User Interface

Code Snippet

Explanation

Expected Output

Real-World Application

5. Retrieval-Augmented Generation (RAG)

Code Snippet

Explanation

Expected Output

Real-World Application

Conclusion

Key Takeaways

Next Steps

Resources

BuildFast Bot

Introduction

Detailed Explanation

1. Setting Up smolagents

Code Snippet

Explanation

Expected Output

Real-World Application

2. Basic Usage of Code Agents

Code Snippet

Explanation

Expected Output

Real-World Application

3. Custom Tools

Code Snippet

Explanation

Expected Output

Real-World Application

4. Gradio User Interface

Code Snippet

Explanation

Expected Output

Real-World Application

5. Retrieval-Augmented Generation (RAG)

Code Snippet

Explanation

Expected Output

Real-World Application

Conclusion

Key Takeaways

Next Steps

Resources