Generative AI Stage Prompt

Prompting stage for generative ai. During this phase, the system collects information from predefined templates and user queries. It then proceeds to replace specific variables denoted as {VARIABLE} within these templates or contextual information.

These values, are determined by the Mapping Stage and are subsequently deposited into the intermediate object. As an example, if you wish to incorporate query results from a previous stage, the Mapping Stage serves to extract these results and organize them into a unified value. This unified value is then passed as an element within the intermediate object.

Alternatively, a personalized stage may be employed in place of the Mapping Stage to assemble these values into a single object within the intermediate structure. However, it is imperative to ensure the meticulous construction of this object to avoid any issues.

Properties

Property	Description	Default	Type	Required
open_ai_use_https	If this flag is True, the stage is going to use the https version of open ai, if not it uses the python version.	False	boolean	No
open_ai_api_key	This is your openai key to use the chat from your service provider.	os.environ.get('OPEN_AI_API_KEY', 'default_key')	string	Yes
open_ai_api_base_url	Base url of your service provider for open ai chat.	os.environ.get('OPEN_AI_API_BASE_URL', 'default_url')	string	Yes
open_ai_type	API type, its restricted by the types allowed by the enum.	EnumOpenAiType.Azure	EnumOpenAiType	Yes
open_ai_api_version	API version of the openai chat endpoint.	os.environ.get('OPEN_AI_API_VERSION', '2023-03-15-preview')	string	Yes
open_ai_engine_name	Name of the engine to be used.	gpt-35-turbo	string	No
open_ai_model_name	Name of the model to be used.	gpt-35-turbo	string	No
open_ai_temperature	Temperature hyperparameter for open ai chat.	0.0	string	No
open_ai_max_tokens	Max tokens hyperparameter for open ai chat.	1500	int	No
open_ai_frequency_penalty	Hyperparameter for open ai chat.	0.0	float	No
open_ai_presence_penalty	Hyperparameter for open ai chat.	0.0	float	No
open_ai_top_p	Hyperparameter for open ai chat.	1.0	float	No
open_ai_certificates_path	Path to pem with the certificates for your open ai chat provider.	None	string	No
prompt_templates_begin	Path to prompt template file to be used as the starting message.	config/prompt_templates/begin/begin.tlp	Path	No
prompt_templates_context	Path to prompt template file to be used as the context message or messages. This parameter is only used when there's no context provided in the intermediate using the api request.	config/prompt_templates/context/context.tlp	Path	No
prompt_templates_query	Path to prompt template file to be used as the question to ask to open ai chat at the current iteration.	config/prompt_templates/query/test.tlp	Path	No
data_preparation_stage_name	Name of the stage where the data is prepared to be ingested in the template.	-	string	Yes
data_preparation_keys	The data preparation keys to be extracted (url, id, doc_title).	"CONTENT:doc_id, content"	Union[string, None]	No
open_ai_url_path	The rest of the url to call open ai when using https, it's added to the api base url. If not provided, the link is builded with the other parameters provided (engine name, type, version, etc)	None	string	No

For more info about open ai hyperparameters check the azure openai documentation Customize a model with Azure OpenAI Service - Azure OpenAI | Microsoft Learn

You can download the certificates from the main url of your service provider, using the browser button to check site's certificates information.

Generative AI Stage Prompt Intermediate Parameters

Parameter	Description
context	This is the context passed on the api call which is going to act as a template instead of the prompt_template_context. We recommend to use this feature instead of the template itself since the context for the prompt may be the previous interactions on a session so the open ai chat can have a record of the conversation it is having with the user.

Remember that the intermediate can be fill with either other stages or the original request body that trigger the pipeline, making this essentially REST API parameters

Example Configuration

_generative_ai_prompt = GenerativeAIPromptStage(
    type='GenerativeAIPromptStage',
    enable=True,
    name='MoviesPrompt',
    open_ai_use_https=True,
    open_ai_api_key='yourapikey',
    open_ai_api_base_url='baseurl',
    open_ai_api_type=EnumOpenAiType.AZURE,
    open_ai_certificates_path='path/to/certs.pem',
    data_preparation_stage_name=_search_mapping.name,
    data_preparation_keys="CONTENT:MOVIE_TITLE, CONTENT"
)

Space shortcuts

Page tree

Properties

Generative AI Stage Prompt Intermediate Parameters

Example Configuration