UNPKG

@n8n/n8n-nodes-langchain

Version:
67 lines (61 loc) • 3.24 kB
/** * DeepSeek Chat Model Node - Version 1 * For advanced usage with an AI chain */ export interface LcLmChatDeepSeekV1Params { /** * The model which will generate the completion. &lt;a href="https://api-docs.deepseek.com/quick_start/pricing"&gt;Learn more&lt;/a&gt;. * @builderHint Choose deepseek-chat or deepseek-reasoner according to the task. Confirm their current capabilities in the provider documentation instead of assuming a fixed underlying version. Keep a working or explicitly requested model unless provider evidence shows it is unavailable or unsuitable. When choosing a new model, select a current stable model supported by this node version, the configured API, and the required operation. Use the declared model-list lookup with the connected credential when available. Gateway credits can restrict that list. For a new choice without credentials, use current provider catalog information. Choose a model that fits the task and budget. Copy the exact model ID. Do not rely on node defaults, invent model IDs, or treat an unfamiliar model as invalid. * @default deepseek-chat */ model?: string | Expression<string>; /** * Additional options to add * @default {} */ options?: { /** 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 * @default 0 */ frequencyPenalty?: number | Expression<number>; /** The maximum number of tokens to generate in the completion. Most models have a context length of 2048 tokens (except for the newest models, which support 32,768). * @default -1 */ maxTokens?: number | Expression<number>; /** Response Format * @default text */ responseFormat?: 'text' | 'json_object' | Expression<string>; /** 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 * @default 0 */ presencePenalty?: number | Expression<number>; /** Controls randomness: Lowering results in less random completions. As the temperature approaches zero, the model will become deterministic and repetitive. * @default 0.7 */ temperature?: number | Expression<number>; /** Maximum amount of time a request is allowed to take in milliseconds * @default 360000 */ timeout?: number | Expression<number>; /** Maximum number of retries to attempt * @default 2 */ maxRetries?: number | Expression<number>; /** Controls diversity via nucleus sampling: 0.5 means half of all likelihood-weighted options are considered. We generally recommend altering this or temperature but not both. * @default 1 */ topP?: number | Expression<number>; }; } export interface LcLmChatDeepSeekV1Credentials { deepSeekApi: CredentialReference; } interface LcLmChatDeepSeekV1NodeBase { type: '@n8n/n8n-nodes-langchain.lmChatDeepSeek'; version: 1; } export type LcLmChatDeepSeekV1ParamsNode = LcLmChatDeepSeekV1NodeBase & { config: NodeConfig<LcLmChatDeepSeekV1Params> & { credentials?: LcLmChatDeepSeekV1Credentials }; }; export type LcLmChatDeepSeekV1Node = LcLmChatDeepSeekV1ParamsNode;