The generated code is then executed to produce the result.

You can instantiate the LLM by passing it as a config to the SmartDataFrame or SmartDatalake constructor.

​

OpenAI models

In order to use OpenAI models, you need to have an OpenAI API key. You can get one here.

Once you have an API key, you can use it to instantiate an OpenAI object:

from pandasai import SmartDataframe
from pandasai.llm import OpenAI

llm = OpenAI(api_token="my-openai-api-key")
pandas_ai = SmartDataframe("data.csv", config={"llm": llm})

As an alternative, you can set the OPENAI_API_KEY environment variable and instantiate the OpenAI object without passing the API key:

from pandasai import SmartDataframe
from pandasai.llm import OpenAI

llm = OpenAI()  # no need to pass the API key, it will be read from the environment variable
pandas_ai = SmartDataframe("data.csv", config={"llm": llm})

If you are behind an explicit proxy, you can specify openai_proxy when instantiating the OpenAI object or set the OPENAI_PROXY environment variable to pass through.

​

Count tokens

You can count the number of tokens used by a prompt as follows:

"""Example of using PandasAI with a pandas dataframe"""

from pandasai import SmartDataframe
from pandasai.llm import OpenAI
from pandasai.helpers.openai_info import get_openai_callback
import pandas as pd

llm = OpenAI()

# conversational=False is supposed to display lower usage and cost
df = SmartDataframe("data.csv", config={"llm": llm, "conversational": False})

with get_openai_callback() as cb:
    response = df.chat("Calculate the sum of the gdp of north american countries")

    print(response)
    print(cb)
#  The sum of the GDP of North American countries is 19,294,482,071,552.
#  Tokens Used: 375
#	Prompt Tokens: 210
#	Completion Tokens: 165
# Total Cost (USD): $ 0.000750

​

Google PaLM

In order to use Google PaLM models, you need to have a Google Cloud API key. You can get one here.

Once you have an API key, you can use it to instantiate a Google PaLM object:

from pandasai import SmartDataframe
from pandasai.llm import GooglePalm

llm = GooglePalm(api_key="my-google-cloud-api-key")
df = SmartDataframe("data.csv", config={"llm": llm})

​

Google Vertexai

In order to use Google PaLM models through Vertexai api, you need to have

1. Google Cloud Project
2. Region of Project Set up
3. Install optional dependency google-cloud-aiplatform
4. Authentication of gcloud

Once you have basic setup, you can use it to instantiate a Google PaLM through vertex ai:

from pandasai import SmartDataframe
from pandasai.llm import GoogleVertexAI

llm = GoogleVertexAI(project_id="generative-ai-training",
                     location="us-central1",
                     model="text-bison@001")
df = SmartDataframe("data.csv", config={"llm": llm})

​

Azure OpenAI

In order to use Azure OpenAI models, you need to have an Azure OpenAI API key as well as an Azure OpenAI endpoint. You can get one here.

To instantiate an Azure OpenAI object you also need to specify the name of your deployed model on Azure and the API version:

from pandasai import SmartDataframe
from pandasai.llm import AzureOpenAI

llm = AzureOpenAI(
    api_token="my-azure-openai-api-key",
    azure_endpoint="my-azure-openai-api-endpoint",
    api_version="2023-05-15",
    deployment_name="my-deployment-name"
)
df = SmartDataframe("data.csv", config={"llm": llm})

As an alternative, you can set the AZURE_OPENAI_API_KEY, OPENAI_API_VERSION, and AZURE_OPENAI_ENDPOINT environment variables and instantiate the Azure OpenAI object without passing them:

from pandasai import SmartDataframe
from pandasai.llm import AzureOpenAI

llm = AzureOpenAI(
    deployment_name="my-deployment-name"
)  # no need to pass the API key, endpoint and API version. They are read from the environment variable
df = SmartDataframe("data.csv", config={"llm": llm})

If you are behind an explicit proxy, you can specify openai_proxy when instantiating the AzureOpenAI object or set the OPENAI_PROXY environment variable to pass through.

​

HuggingFace via Text Generation

In order to use HuggingFace models via text-generation, you need to first serve a supported large language model (LLM). Read text-generation docs for more on how to setup an inference server.

This can be used, for example, to use models like LLaMa2, CodeLLaMa, etc. You can find more information about text-generation here.

The inference_server_url is the only required parameter to instantiate an HuggingFaceTextGen model:

from pandasai.llm import HuggingFaceTextGen
from pandasai import SmartDataframe

llm = HuggingFaceTextGen(
    inference_server_url="http://127.0.0.1:8080"
)
df = SmartDataframe("data.csv", config={"llm": llm})

​

LangChain models

PandasAI has also built-in support for LangChain models.

In order to use LangChain models, you need to install the langchain package:

pip install pandasai[langchain]

Once you have installed the langchain package, you can use it to instantiate a LangChain object:

from pandasai import SmartDataframe
from langchain_openai import OpenAI

langchain_llm = OpenAI(openai_api_key="my-openai-api-key")
df = SmartDataframe("data.csv", config={"llm": langchain_llm})

PandasAI will automatically detect that you are using a LangChain LLM and will convert it to a PandasAI LLM.

​

Amazon Bedrock models

In order to use Amazon Bedrock models, you need to have an AWS AKSK and gain the model access.

Currently, only Claude 3 Sonnet is supported.

In order to use Bedrock models, you need to install the bedrock package.

pip install pandasai[bedrock]

Then you can use the Bedrock models as follows

from pandasai import SmartDataframe
from pandasai.llm import BedrockClaude
import boto3

bedrock_runtime_client = boto3.client(
    'bedrock-runtime',
    aws_access_key_id=ACCESS_KEY,
    aws_secret_access_key=SECRET_KEY
)

llm = BedrockClaude(bedrock_runtime_client)
df = SmartDataframe("data.csv", config={"llm": llm})

More ways to create the bedrock_runtime_client can be found here.

​

More information

For more information about LangChain models, please refer to the LangChain documentation.

​

IBM watsonx.ai models

In order to use IBM watsonx.ai models, you need to have

1. IBM Cloud api key
2. Watson Studio project in IBM Cloud
3. The service URL associated with the project’s region

The api key can be created in IBM Cloud. The project ID can determined after a Watson Studio service is provisioned in IBM Cloud. The ID can then be found in the project’s Manage tab (Project -> Manage -> General -> Details). The service url depends on the region of the provisioned service instance and can be found here.

In order to use watsonx.ai models, you need to install the ibm-watsonx-ai package.

At this time, watsonx.ai does not support the PandasAI agent.

pip install pandasai[ibm-watsonx-ai]

Then you can use the watsonx.ai models as follows

from pandasai import SmartDataframe
from pandasai.llm import IBMwatsonx

llm = IBMwatsonx(
    model="ibm/granite-13b-chat-v2",
    api_key=API_KEY,
    watsonx_url=WATSONX_URL,
    watsonx_project_id=PROJECT_ID,
)

df = SmartDataframe("data.csv", config={"llm": llm})

​

More information

For more information on the watsonx.ai SDK you can read more here.

​

Local models

PandasAI supports local models, though smaller models typically don’t perform as well. To use local models, first host one on a local inference server that adheres to the OpenAI API. This has been tested to work with Ollama and LM Studio.

​

Ollama

Ollama’s compatibility is experimental (see docs).

With an Ollama server, you can instantiate an LLM object by specifying the model name:

from pandasai import SmartDataframe
from pandasai.llm.local_llm import LocalLLM

ollama_llm = LocalLLM(api_base="http://localhost:11434/v1", model="codellama")
df = SmartDataframe("data.csv", config={"llm": ollama_llm})

​

LM Studio

An LM Studio server only hosts one model, so you can instantiate an LLM object without specifying the model name:

from pandasai import SmartDataframe
from pandasai.llm.local_llm import LocalLLM

lm_studio_llm = LocalLLM(api_base="http://localhost:1234/v1")
df = SmartDataframe("data.csv", config={"llm": lm_studio_llm})