aiaas-falcon

AIaaS Falcon

Installation | Quickstart |

Description

AIaaS_Falcon is Generative AI - LLM library interacts with different model api endpoints., allowing operations such as listing models, creating embeddings, and generating text based on certain configurations. AIaaS_Falcon helps to invoking the RAG pipeline in seconds.

Supported Endpoint Types:

• Azure OpenAI
• SingtelGPT
• Dev_Quantized
• Dev_Full

:shield: Installation

Ensure you have the requests and google-api-core libraries installed:

pip install aiaas-falcon

if you want to install from source

git clone https://github.com/Praveengovianalytics/AIaaS_falcon && cd AIaaS_falcon
pip install -e .

Methods

Falcon Class

• __init__ (config) Intialise the Falcon object with endpoint configs.
  Parameter:

  • api_key : API Key
  • api_name: Name for endpoint
  • api_endpoint: Type of endpoint ( can be azure, dev_quan, dev_full, prod)
  • host_name_port: Host and Port Inforation
  • use_pil: Activate Personal Identifier Information Limit Protection (Boolean)
  • protocol: HTTP/ HTTPS
  • api_type: Subroute if needed
  • use_pii: Whether current endpoint need Personal Identifier Information Limit Protection
  • log_key: Auth Key to use the Application

• current_active() Check current endpoint active

• add_endpoint(api_name,protocol,host_name_port,api_endpoint,api_key,use_pil=False) Add a new endpoint.
  Parameter:

  • api_key : API Key
  • api_name: Name for endpoint
  • api_endpoint: Type of endpoint ( can be azure, dev_quan, dev_full, prod)
  • host_name_port: Host and Port Inforation
  • use_pii: Activate Personal Identifier Information Limit Protection (Boolean)
  • protocol: HTTP/ HTTPS
  • use_pil: Whether current endpoint need Personal Identifier Information Limit Protection

• list_endpoint() List out all endpoints in the endpoint management manager

• set_endpoint(name) Set target endpoint to active
  Parameter:

  • name : Target endpoint's name

• remove_endpoint(name) Delete endpoint by name
  Parameter:

  • name : Target endpoint's name

• current_pii() Check current Personal Identifier Information Protection activation status

• switch_pii() Switch current Personal Identifier Information Protection activation status

• list_models() List out models available

• initalise_pii() Download and intialise PII Protection.
  Note: This does not activate PII but initialise dependencies

• health() Check health of current endpoint

• create_embedding(file_path) Create embeddings by sending files to the API.
  Parameter:

  • file_path: Path to file

• generate_text_full(query="", context="", use_file=0, model="", chat_history=[], max_new_tokens: int = 200, temperature: float = 0, top_k: int = -1, frequency_penalty: int = 0, repetition_penalty: int = 1, presence_penalty: float = 0, fetch_k=100000, select_k=4, api_version='2023-05-15', guardrail={'jailbreak': False, 'moderation': False}, custom_guardrail=None)
  Generate text using LLM endpoint. Note: Some parameter of the endpoint is endpoint-specific.
  Parameter:

  • query: a string of your prompt
  • use_file: Whether to take file to context in generation. Only applies to dev_full and dev_quan. Need to create_embedding before use.
  • model: a string on the model to use. You can use list_models to check for model available.
  • chat_history: an array of chat history between user and bot. Only applies to dev_full and dev_quan. (Beta)
  • max_new_token: maximum new token to generate. Must be integer.
  • temperature: Float that controls the randomness of the sampling. Lower values make the model more deterministic, while higher values make the model more random. Zero means greedy sampling.
  • top_k: Integer that controls the number of top tokens to consider.
  • frequency_penalty: Float that penalizes new tokens based on their frequency in the generated text so far.
  • repetition_penalty: Float that penalizes new tokens based on whether they appear in the prompt and the generated text so far.
  • presence_penalty: Float that penalizes new tokens based on whether they appear in the generated text so far
  • fetch_k: Use for document retrival. Include how many element in searching. Only applies when use_file is 1
  • select k: Use to select number of document for document retrieval. Only applies when use_file is 1
  • api_version: Only applies for azure endpoint
  • guardrail: Whether to use the default jailbreak guardrail and moderation guardrail
  • custom_guardrail: Path to custom guardrail .yaml file. The format can be found in sample.yaml

• evaluate_parameter(config) Carry out grid search for parameter
  Parameter:

  • config: A dict. The dict must contain model and query. Parameter to grid search must be a list.
    • model: a string of model
    • query: a string of query
    • **other parameter (eg: "temperature":list(np.arange(0,2,0.5))

• decrypt_hash(encrypted_data) Decret the configuration from experiment id. Parameter:

  • encrypted_data: a string of id

:fire: Quickstart

from aiaas_falcon import Falcon
model=Falcon(api_name="azure_1",protocol='https',host_name_port='example.com',api_key='API_KEY',api_endpoint='azure',log_key="KEY")
model.list_models()
model.generate_text_full(query="Hello, introduce yourself",model='gpt-35-turbo-0613-vanilla',api_version='2023-05-15')

Conclusion

AIaaS_Falcon library simplifies interactions with the LLM API's, providing a straightforward way to perform various operations such as listing models, creating embeddings, and generating text.

Authors

• @Praveengovianalytics
• @zhuofan

Google Colab

• Get start with aiaas_falcon

Badges