Created
August 30, 2023 11:14
-
-
Save Necmttn/5224c7968a58ea406eca37ab1f1b2d29 to your computer and use it in GitHub Desktop.
Stream execute openai call including the JSON responses.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import { | |
createParser, | |
ParsedEvent, | |
ReconnectInterval, | |
} from "eventsource-parser"; | |
import { Procedure } from "./procedure"; | |
import openai, { ChatCompletionRequestMessage } from "@acme/openai"; | |
import _ from "lodash"; | |
export type Message = Omit<ChatCompletionRequestMessage, "role"> & { | |
role: "user" | "assistant" | "function" | "system"; | |
name?: string; | |
content: string | null; | |
function_call?: { | |
name: string; | |
arguments: string; | |
}; | |
}; | |
type GPT_MODELS = | |
| "gpt-3.5-turbo-0613" | |
| "gpt-3.5-turbo-16k-0613" | |
| "gpt-4-0613"; | |
export async function getApiResponse({ | |
history, | |
procedures, | |
model = "gpt-3.5-turbo-0613", | |
stream = true, | |
}: { | |
model?: GPT_MODELS; | |
history: Message[]; | |
procedures: Procedure<any, any, any>[]; | |
stream?: boolean; | |
}) { | |
console.log("@".repeat(60)); | |
console.log({ | |
messages: _.last(history), | |
}); | |
const response = await openai.createChatCompletion({ | |
messages: history, | |
model, | |
stream, | |
temperature: 0.4, | |
functions: | |
procedures.length > 0 ? procedures.map((p) => p.toObject()) : undefined, | |
}); | |
return response; | |
} | |
async function executeProcedure( | |
procedures: Procedure<any, any, any>[], | |
functionName: string, | |
args: string, | |
ctx: any, | |
) { | |
const procedure = procedures.find((p) => p.name === functionName); | |
if (!procedure) { | |
throw new Error(`Unknown function call: ${functionName}`); | |
} | |
const parsedArguments = JSON.parse(args); | |
const result = await procedure.operation({ | |
ctx, | |
input: parsedArguments, | |
}); | |
return result; | |
} | |
function mergeObjects<T extends { [key: string]: any }>(objects: T[]): T { | |
return objects.reduce((accumulator, current) => { | |
for (const key in current) { | |
if (current.hasOwnProperty(key)) { | |
if ( | |
typeof current[key] === "string" && | |
typeof accumulator[key] === "string" | |
) { | |
accumulator[key] = ((accumulator[key] as string) + | |
(current[key] as string)) as any; | |
} else if (current[key] !== undefined) { | |
if ( | |
typeof current[key] === "object" && | |
!Array.isArray(current[key]) && | |
current[key] !== null | |
) { | |
// If it's a nested object, merge it | |
accumulator[key] = mergeObjects([ | |
accumulator[key] || {}, | |
current[key], | |
]) as any; | |
} else { | |
accumulator[key] = current[key]; | |
} | |
} | |
} | |
} | |
return accumulator; | |
}, {} as T); | |
} | |
export async function executeStreamProcedures({ | |
model, | |
history, | |
procedures, | |
ctx, | |
onResponse, | |
}: { | |
model?: GPT_MODELS; | |
history: Message[]; | |
procedures: Procedure<any, any, any>[]; | |
ctx: any; | |
onResponse: (content: string) => Promise<void>; | |
}) { | |
const encoder = new TextEncoder(); | |
const decoder = new TextDecoder(); | |
const stream = new ReadableStream({ | |
async start(controller) { | |
let latestResponse: Partial<{ | |
index: number; | |
delta: { | |
role: string; | |
function_call?: { | |
name: string; | |
arguments: string; | |
}; | |
content: string | null; | |
}; | |
finish_reason: "stop" | "function_call" | null; | |
}> = {}; | |
async function onParse(event: ParsedEvent | ReconnectInterval) { | |
if (event.type === "event") { | |
const data = event.data; | |
if (data === "[DONE]") { | |
if (latestResponse?.finish_reason === "stop") { | |
console.log("Its finished."); | |
await onResponse(latestResponse?.delta?.content!); | |
controller.close(); | |
return; | |
} | |
} else { | |
try { | |
const json = JSON.parse(data); | |
if (json?.error) { | |
throw new Error(json?.error); | |
} | |
const message = json.choices[0]; | |
latestResponse = mergeObjects([latestResponse, message]); | |
const text = message?.delta.content; | |
if (text) { | |
const queue = encoder.encode(text); | |
controller.enqueue(queue); | |
} | |
} catch (e) { | |
console.log(e); | |
controller.error(e); | |
} | |
} | |
} | |
} | |
const parser = createParser(onParse); | |
while (true) { | |
const response = await getApiResponse({ history, procedures, model }); | |
for await (const chunk of response.body as any) { | |
parser.feed(decoder.decode(chunk)); | |
} | |
if (!!latestResponse.index) { | |
console.log("EMPTY RESPONSE"); | |
break; | |
} | |
console.log("LATEST", latestResponse); | |
if (latestResponse?.delta?.function_call) { | |
history.push({ | |
role: "assistant", | |
content: latestResponse?.delta?.content!, | |
function_call: latestResponse?.delta?.function_call, | |
}); | |
const procedureResult = await executeProcedure( | |
procedures, | |
latestResponse?.delta?.function_call.name, | |
latestResponse.delta?.function_call.arguments, | |
ctx, | |
); | |
history.push({ | |
role: "function", | |
name: latestResponse.delta?.function_call.name, | |
content: JSON.stringify(procedureResult), | |
}); | |
latestResponse = {}; | |
continue; | |
} | |
if (latestResponse?.delta?.content) { | |
history.push({ | |
role: "assistant", | |
content: latestResponse?.delta?.content as string, | |
}); | |
console.log({ | |
"GOT RESPONSEEE": latestResponse?.delta?.content, | |
}); | |
// await onResponse(latestResponse?.delta?.content); | |
} | |
if (latestResponse.finish_reason === "stop") { | |
break; | |
} | |
if (!latestResponse) { | |
console.log("SOMETHING WENT SOUTH."); | |
break; | |
} | |
} | |
}, | |
}); | |
return stream; | |
} |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment