LlamaIndex LLM Integration: Aleph Alpha

This README details the process of integrating Aleph Alpha's Large Language Models (LLMs) with LlamaIndex. Utilizing Aleph Alpha's API, users can generate completions, facilitate question-answering, and perform a variety of other natural language processing tasks directly within the LlamaIndex framework.

Features

• Text Completion: Use Aleph Alpha LLMs to generate text completions for prompts.
• Model Selection: Access the latest Aleph Alpha models, including the Luminous model family, to generate responses.
• Advanced Sampling Controls: Customize the response generation with parameters like temperature, top_k, top_p, presence_penalty, and more, to fine-tune the creativity and relevance of the generated text.
• Control Parameters: Apply attention control parameters for advanced use cases, affecting how the model focuses on different parts of the input.

Installation

pip install llama-index-llms-alephalpha

Usage

from llama_index.llms.alephalpha import AlephAlpha

1. Request Parameters:

   • model: Specify the model name (e.g., luminous-base-control). The latest model version is always used.
   • prompt: The text prompt for the model to complete.
   • maximum_tokens: The maximum number of tokens to generate.
   • temperature: Adjusts the randomness of the completions.
   • top_k: Limits the sampled tokens to the top k probabilities.
   • top_p: Limits the sampled tokens to the cumulative probability of the top tokens.
   • log_probs: Set to true to return the log probabilities of the tokens.
   • echo: Set to true to return the input prompt along with the completion.
   • penalty_exceptions: A list of tokens that should not be penalized.
   • n: Number of completions to generate.

2. Advanced Sampling Parameters: (Optional)

   • presence_penalty & frequency_penalty: Adjust to discourage repetition.
   • sequence_penalty: Reduces likelihood of repeating token sequences.
   • hosting: Option to process the request in Aleph Alpha's own datacenters for enhanced data privacy.

Response Structure

* `model_version`: The name and version of the model used.
* `completions`: A list containing the generated text completion(s) and optional metadata:
    * `completion`: The generated text completion.
    * `log_probs`: Log probabilities of the tokens in the completion.
    * `raw_completion`: The raw completion without any post-processing.
    * `completion_tokens`: Completion split into tokens.
    * `finish_reason`: Reason for completion termination.
* `num_tokens_prompt_total`: Total number of tokens in the input prompt.
* `num_tokens_generated`: Number of tokens generated in the completion.

Example

Refer to the example notebook for a comprehensive guide on generating text completions with Aleph Alpha models in LlamaIndex.

API Documentation

For further details on the API and available models, please consult Aleph Alpha's API Documentation.