sanity checking, do not duplicate responses
This commit is contained in:
		
							
								
								
									
										16
									
								
								types.d.ts
									
									
									
									
										vendored
									
									
								
							
							
						
						
									
										16
									
								
								types.d.ts
									
									
									
									
										vendored
									
									
								
							| @ -69,15 +69,28 @@ export interface Mention { | ||||
|   username: string; | ||||
| } | ||||
|  | ||||
| /** | ||||
|  * Experimental settings, I wouldn't recommend messing with these if you don't know how they work (I don't either) | ||||
|  */ | ||||
| export interface OllamaConfigOptions { | ||||
|   /** | ||||
|    * Number of tokens guaranteed to be kept in memory during response generation. Higher values leave less | ||||
|    * possible room for num_ctx | ||||
|    */ | ||||
|   num_keep?: number; | ||||
|   seed?: number; | ||||
|   /** | ||||
|    * Sets maximum of tokens in the response | ||||
|    */ | ||||
|   num_predict?: number; | ||||
|   top_k?: number; | ||||
|   top_p?: number; | ||||
|   min_p?: number; | ||||
|   typical_p?: number; | ||||
|   repeat_last_n?: number; | ||||
|   /** | ||||
|    * How close of a response should the response be to the original prompt - lower = more focused response | ||||
|    */ | ||||
|   temperature?: number; | ||||
|   repeat_penalty?: number; | ||||
|   presence_penalty?: number; | ||||
| @ -88,6 +101,9 @@ export interface OllamaConfigOptions { | ||||
|   penalize_newline?: boolean; | ||||
|   stop?: string[]; | ||||
|   numa?: boolean; | ||||
|   /** | ||||
|    * Number of tokens for the prompt to keep in memory for the response, minus the value of num_keep | ||||
|    */ | ||||
|   num_ctx?: number; | ||||
|   num_batch?: number; | ||||
|   num_gpu?: number; | ||||
|  | ||||
		Reference in New Issue
	
	Block a user