Add chat-plugin #134
Add chat-plugin #134
Conversation
…rk on imbedding open ai calls
…ide of data collection features
… that triggers after a certain number of messages or response time prompting the user and creating a continue button
…ing and continue button
…s for customizing the flow of conversation
…work for everything except chainPrompts
… chat with no title
…and fixing data collection to be more legible
🦋 Changeset detectedLatest commit: 80b5c60 The changes in this PR will be included in the next version bump. This PR includes changesets to release 1 package
Not sure what this means? Click here to learn what changesets are. Click here if you're a maintainer who wants to add another changeset to this PR |
|
Hey @vzhang03 and @Bankminer78, any interest in updating this for version 8? Have a student project that wants to use it. I can work on it soon if neither of you has the time right now 😁 |
| @@ -41,6 +41,13 @@ const info = <const>{ | |||
| type: ParameterType.STRING, | |||
| default: null, | |||
| }, | |||
| /** | |||
There was a problem hiding this comment.
artifact from a previous commit?
|
|
||
| | Parameter | Type | Default Value | Description | | ||
| | ------------------- | ---------------- | ------------------ | ---------------------------------------- | | ||
| | ai_prompt | String | undefined | System prompt that will be fed to GPT in order to prompt the chatbot. This is the most simple prompting parameter and should be used in cases where you want to | |
There was a problem hiding this comment.
not finished description
| | ai_prompt | String | undefined | System prompt that will be fed to GPT in order to prompt the chatbot. This is the most simple prompting parameter and should be used in cases where you want to | | ||
| | ai_model | String | gpt-4o-mini | Refers to the OpenAI model that will be called during the chats. | | ||
| | continue_button | Object | { message_trigger: 0 } | Allows you to customize when the continue_button appears. The three options that can be passed in the object are "timer_trigger", "message_trigger", and "message". "Timer_trigger" referes to the milliseconds from initialization, "message_trigger" refers to the total number of messages sent by the user and "message" is the message to be displayed when the button appears. This is an optional parameter that should be used to customize the length of conversation. | | ||
| | ai_model | String | gpt-4o-mini | OpenAI model that will be called during the chats. | |
|
|
||
| jest.useFakeTimers(); | ||
|
|
||
| describe("my plugin", () => { |
There was a problem hiding this comment.
guessing that it would be really hard to mock this sort of plugin, so best to just remove the test and update the package.json respectively
| const info = <const>{ | ||
| name: "chat", | ||
| parameters: { | ||
| // BOOL, STRING, INT, FLOAT, FUNCTION, KEY, KEYS, SELECT, HTML_STRING, IMAGE, AUDIO, VIDEO, OBJECT, COMPLEX |
There was a problem hiding this comment.
best to TSDoc these as well for developer ease of use
| // type: ParameterType.STRING, | ||
| // default: undefined, | ||
| // }, | ||
| continue_button: { |
There was a problem hiding this comment.
is it worth grouping all of these? or can we just have three separate parameters?
| array: true, | ||
| default: [], | ||
| }, | ||
| selection_prompt: { |
There was a problem hiding this comment.
selection_prompt: { selection_prompt: ... } feels a bit unwieldy, maybe selection_stimulus?
| }, | ||
| message_trigger: { | ||
| type: ParameterType.INT, | ||
| default: 99999999999999999999999, // silencing error message |
There was a problem hiding this comment.
possible to -1? not sure why this would need one when the others don't especially if prompt_chain's default is {}
| trial.selection_prompt["message_trigger"] === null && | ||
| trial.selection_prompt["timer_trigger"] === null | ||
| ) { | ||
| console.error("Missing required trigger property in selection_prompt, will never trigger"); |
There was a problem hiding this comment.
style, but best to use throw new Error("....") to halt execution
|
|
||
| const system_user = | ||
| this.selection_prompt["selection_prompt"] + | ||
| "Task: You should select one of the three possible responses that would best achieve your goal with the intention of outputting it to the user" + |
There was a problem hiding this comment.
possible to parameterize this?
Wanted to submit this PR to get a review on what changes should be worked on next. The functionality is all working as intended and has gone through testing by fellow URSI team. This plugin contains functionality to customize a chat with an LLM supporting various strategies for controlling bot-prompting and user prompting. This plugin also collects user keystrokes and comprehensive data on the conversation and communication with the LLM.
Main thing to work on is documentation and specifically how to guide people to navigate the process of using the backend. This plugin is hard to work with for people without coding experience, but not sure how to remedy this issue.
A smaller issue is a code rewrite moving the fetching and calling into a separate class. This would make it more friendly for devs to work with. Haven't done because prioritized building out prompting features for the other group. Not sure if necessary or if should just comment on the associated methods.
Lastly, should figure out where to handle styling and if should move away from
styles.cssto another place as well as allow researcher to customize.