ai21 1.1.3

No project description provided

AI21 Studio Client

Python client for the AI21 Studio API

API Keys

API key can be set globally, using ai21.api_key, or alternatively - be passed to any function call as a named argument. For example - ai21.Dataset.get(api_key=api_key)

Installation:

pip install -U ai21

Client customization

The ai21 namespace has parameters that one can set globally, and they will be used by all the following client calls. One can also pass each of these parameters as a named argument. If a parameter is overridden globally, and also passed as a named argument - the passed named argument will be used with the following client calls. The supported parameters are described below:

* api_key (str) - the key to be used for the Authorization header
* organization (str) - this information is sent as part of the user-agent
* application (str) - this information is sent as part of the user-agent
* api_version (str) - the api version. (i.e. v1 / v2). Default=v1
* api_host (str) - the api host. Default=https://api.ai21.com
* timeout_sec (int) - the number of seconds the client must wait before cutting the connection. Default=30
* num_retries (int) - the number of attempts to retry a failure response. Default=0 (no retries)
* log_level (str) - debug / info / error (Default=error)

Retry on Failure

By default, there is no retry on failures. If setting the num_retries parameter to an integer greater than 0, then all the following client calls will retry up to the specified number of times on the following http error codes: 429, 500, 503.

Documentation & resources

https://docs.ai21.com/

Usage

Studio API

In the following example, an AI21 API client is used with an API key and a timeout parameter set globally. This example uses a simple completion call with only prompt and maxTokens arguments supplied (all other completion settings are set to their defaults):

import ai21

ai21.api_key = 'my_api_key'
ai21.timeout_sec = 20
response = ai21.Completion.execute(model="j2-ultra", prompt="Write a news release in the voice of a global banking conglomerate announcing an unprecedented building campaign to expand and rebuild their corporate headquarters in Alpha Centauri.", maxTokens=200)
print(response)

AWS Client

This python SDK can also be used to invoke Jurassic models as a SageMaker (SM) endpoint or as an AWS Bedrock API. To activate this option, make sure to install with the extra AWS dependencies:

pip install -U "ai21[AWS]"

Sagemaker endpoint request

You can then generate a completion by running:

import ai21
ai21.aws_region = 'us-east-1'
response = ai21.Completion.execute(
    destination=ai21.SageMakerDestination("<your_endpoint_name>"),
    prompt="hello world",
    maxTokens=20,
)

print(response)

Bedrock API request

You can use one of the models offered in the following list as model_id:

• ai21.j2-grande-instruct
• ai21.j2-jumbo-instruct

import ai21

response = ai21.Completion.execute(
    destination=ai21.BedrockDestination(model_id=ai21.BedrockModelID.J2_JUMBO_INSTRUCT),
    prompt="hello world",
    maxTokens=20,
)

print(response)