@isdk/ai-tool-prompt

ai-tool-prompt

• Management of various prompts
• System template format prompts (guessing corresponding system templates based on the model)
• Recommendation of parameters for LLM models

Installation

npm install @isdk/ai-tool @isdk/ai-tool-prompt

Prompt Structure

• _id: string (unique) - id of the prompt
• templateFormat: 'hf'|'ollama'|'langchain' - template format of the prompt, defaults to 'hf'
• template: string - template of the prompt, eg, '{{messages[0].content}}'
• type: 'system'|'tool'|'char' - type of the prompt
• prompt: object - default prompt data
• rule: string|RegExpr|(string|RegExpr)[] - rule to match the LLM model name
• parameters: {[ruleMatched: string]: object]} - the default parameters of LLM models
  • the key is the matched rule named group: id or group[1], it is lowercase always.
  • the value is parameters of the LLM model
• extends: string - the id of the parent prompt
• version: {[name: string]: object} - the version/feature info of the prompt
  • the key is the name of the version/feature
  • the value is the prompt to overwrite.

Changelog

0.0.2 (2024-06-08)

Features

• add $getDefaultPrompt() method (15e161f)
• add $getParameters to get LLM default parameters if any (7066928)
• add all matched prompts (9418140)
• add array supports to registerYamlTag (644b1d2)
• add config init (880729f)
• add const AIPromptsName (ff19eb2)
• add extends support for prompt (d884979)
• add filter the prompt type to $getPrompt (139c47a)
• add formatPrompt func (9006544)
• add input, output to AIPromptSettings (6c07ed4)
• add isFitForLLM method (b42af0a)
• add priority support for prompt (6314dec)
• add registerYamlTag func (f47990e)
• AIPrompt: add isFitForLLM method to check the prompt whether good fit LLM (f49d303)
• AIPrompts: add custom method getPrompt to get a better prompt for LLM (9d91a57)
• AIPromptSettings: add optional contentType field (4495900)
• export ConfigFile (91a8c8c)
• export parseYaml (32b70aa)
• more system prompts template added (8549f34)
• promptIsFitForLLM: the modelName argument can be prompt id now (36e5fbf)
• prompts: $getDefaultPrompt can pass type now (a9363d4)
• prompts: add codeqwen support to ChatML (a8f6601)
• prompts: add Deepseek system prompts (355a50e)
• prompts: add default prompt (5159ba3)
• return all fits for LLM (2acaf91)

Bug Fixes

• add ineractive mode supports (54e00ec)
• AIPromptsFunc: extends should concat array(messages) (4e6041b)
• AIPromptsFunc: get should extends array(concat) too (f4454f0)
• build commonjs need import.meta.url shims (8a6dc4a)
• build: export commonjs and esm (0936b7f)
• can not getPrompt by LLM (7e34c74)
• ChatML: the gguf file build-in eot_token is wrong, so correct it here (260d54c)
• default prompt template (367b858)
• forget to add new options to AIPromptSchema (11bdf21)
• OpenChat: use eot_token instead of end_of_turn (10bc4a0)
• parmeters rule matched name should be lowercase always (13c9772)
• Phi-2: forget to add rule (d993921)
• prompt: add stop words to PHi-3 (59f7920)
• prompt: add stop_words for qwen system prompt template (a499d25)
• prompt: add stop_words to llama3 (90ab21f)
• prompt: for bagel (1ba0c0b)
• promptIsFitForLLM return not matched prompt (7a655b4)
• prompt: let openbuddy embeded system prompt template first (b16b423)
• prompts: add iteractive mode supports (c2ac49e)
• prompts: many changes (2aee0c7)
• prompts: update phi-3 rule (75354f9)
• prompt: with line break (abed8df)
• ts: add_generation_prompt should be a boolean (71007a7)
• ts: treat id as any (4469376)