Files
automatisch/packages/backend/src/apps/self-hosted-llm/actions/send-prompt/index.ts

105 lines
3.5 KiB
TypeScript

import defineAction from '../../../../helpers/define-action';
const castFloatOrUndefined = (value: string | null) => {
return value === '' ? undefined : parseFloat(value);
}
export default defineAction({
name: 'Send prompt',
key: 'sendPrompt',
description: 'Creates a completion for the provided prompt and parameters.',
arguments: [
{
label: 'Model',
key: 'model',
type: 'dropdown' as const,
required: true,
variables: true,
source: {
type: 'query',
name: 'getDynamicData',
arguments: [
{
name: 'key',
value: 'listModels',
},
],
},
},
{
label: 'Prompt',
key: 'prompt',
type: 'string' as const,
required: true,
variables: true,
description: 'The text to analyze.'
},
{
label: 'Temperature',
key: 'temperature',
type: 'string' as const,
required: false,
variables: true,
description: 'What sampling temperature to use. Higher values mean the model will take more risk. Try 0.9 for more creative applications, and 0 for ones with a well-defined answer. We generally recommend altering this or Top P but not both.'
},
{
label: 'Maximum tokens',
key: 'maxTokens',
type: 'string' as const,
required: false,
variables: true,
description: 'The maximum number of tokens to generate in the completion.'
},
{
label: 'Stop Sequence',
key: 'stopSequence',
type: 'string' as const,
required: false,
variables: true,
description: 'Single stop sequence where the API will stop generating further tokens. The returned text will not contain the stop sequence.'
},
{
label: 'Top P',
key: 'topP',
type: 'string' as const,
required: false,
variables: true,
description: 'An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with Top P probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.'
},
{
label: 'Frequency Penalty',
key: 'frequencyPenalty',
type: 'string' as const,
required: false,
variables: true,
description: `Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.`
},
{
label: 'Presence Penalty',
key: 'presencePenalty',
type: 'string' as const,
required: false,
variables: true,
description: `Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.`
},
],
async run($) {
const payload = {
model: $.step.parameters.model as string,
prompt: $.step.parameters.prompt as string,
temperature: castFloatOrUndefined($.step.parameters.temperature as string),
max_tokens: castFloatOrUndefined($.step.parameters.maxTokens as string),
stop: ($.step.parameters.stopSequence as string || null),
top_p: castFloatOrUndefined($.step.parameters.topP as string),
frequency_penalty: castFloatOrUndefined($.step.parameters.frequencyPenalty as string),
presence_penalty: castFloatOrUndefined($.step.parameters.presencePenalty as string),
};
const { data } = await $.http.post('/v1/completions', payload);
$.setActionItem({
raw: data,
});
},
});