LlamaIndex Llms Integration: Mymagic

Installation

To install the required package, run:

bash

%pip install llama-index-llms-mymagic
!pip install llama-index

Setup

Before you begin, set up your cloud storage bucket and grant MyMagic API secure access. For detailed instructions, visit the MyMagic documentation.

Initialize MyMagicAI

Create an instance of MyMagicAI by providing your API key and storage configuration:

python

from llama_index.llms.mymagic import MyMagicAI

llm = MyMagicAI(
    api_key="your-api-key",
    storage_provider="s3",  # Options: 's3' or 'gcs'
    bucket_name="your-bucket-name",
    session="your-session-name",  # Directory for batch inference
    role_arn="your-role-arn",
    system_prompt="your-system-prompt",
    region="your-bucket-region",
    return_output=False,  # Set to True to return output JSON
    input_json_file=None,  # Input file stored on the bucket
    list_inputs=None,  # List of inputs for small batch
    structured_output=None,  # JSON schema of the output
)

Note: If return_output is set to True, max_tokens should be at least 100.

Generate Completions

To generate a text completion for a question, use the complete method:

python

resp = llm.complete(
    question="your-question",
    model="choose-model",  # Supported models: mistral7b, llama7b, mixtral8x7b, codellama70b, llama70b, etc.
    max_tokens=5,  # Number of tokens to generate (default is 10)
)
print(
    resp
)  # The response indicates if the final output is stored in your bucket or raises an exception if the job failed

Asynchronous Requests

For asynchronous operations, use the acomplete endpoint:

python

import asyncio


async def main():
    response = await llm.acomplete(
        question="your-question",
        model="choose-model",  # Supported models listed in the documentation
        max_tokens=5,  # Number of tokens to generate (default is 10)
    )
    print("Async completion response:", response)


await main()

LLM Implementation example

https://docs.llamaindex.ai/en/stable/examples/llm/mymagic/