Skip to content

Commit

Permalink
yolo
Browse files Browse the repository at this point in the history
  • Loading branch information
@Ataxia123
Ataxia123 committed Apr 23, 2024
1 parent 835cc26 commit 985f59c
Show file tree
Hide file tree
Showing 4 changed files with 170 additions and 168 deletions.
313 changes: 155 additions & 158 deletions packages/nextjs/app/ai/llmCall.ts
View file
Original file line number Diff line number Diff line change
Expand Up @@ -4,189 +4,186 @@ import { storePrompt } from "./metrics/prompt";
import { storeUsageByEmbeddingId } from "./metrics/usage";
// Assumed environme
import {
ChatMessage,
Document,
MongoDBAtlasVectorSearch,
VectorStoreIndex,
storageContextFromDefaults,
ChatMessage,
Document,
MongoDBAtlasVectorSearch,
VectorStoreIndex,
storageContextFromDefaults,
} from "llamaindex";
import { OpenAI, serviceContextFromDefaults } from "llamaindex";
import { Db, MongoClient, ObjectId } from "mongodb";
import { Db, MongoClient } from "mongodb";

// Assuming we've defined or imported types for the Hackathon Application

export type Haikipu = {
id: string,
address: string,
title: string,
type: string,
timestamp: string,
contextSummary: any,
haiku: string,
explainer: string,
id: string;
address: string;
title: string;
type: string;
timestamp: string;
contextSummary: any;
haiku: string;
explainer: string;
};

const url = process.env.MONGODB_URL || 'mongodb+srv://At0x:[email protected]/?retryWrites=true&w=majority'
const url =
process.env.MONGODB_URL ||
"mongodb+srv://At0x:[email protected]/?retryWrites=true&w=majority";

const client = new MongoClient(url);
await client.connect();
// Database Name

async function llamaindex(payload: string, id: string) {
const vectorStore = new MongoDBAtlasVectorSearch({
mongodbClient: client,
dbName: "nerdWorkState",
collectionName: "nerdIndex", // this is where your embeddings will be stored
indexName: "nerd_index", // this is the name of the index you will need to create
});

// now create an index from all the Documents and store them in Atlas
const storageContext = await storageContextFromDefaults({ vectorStore });

const essay = payload;

// Create Document object with essay
const document = new Document({ text: essay, id_: id });
console.log({ document });
// Split text and create embeddings. Store them in a VectorStoreIndex
const result = await VectorStoreIndex.fromDocuments([document], { storageContext });
const embeddingResults = await result.getNodeEmbeddingResults([document]);
console.log({ result, embeddingResults });
const db = client.db("nerdWorkState"); // Connect to the database
const hackIndex = db.collection("nerdIndex");

const embedding = await hackIndex.findOne({ "metadata.doc_id": id });

console.log({ embeddingId: embedding?.id });
console.log(`Successfully created embeddings in the MongoDB collection`);
return { embeddingId: embedding?.id as string, result: embeddingResults };
const vectorStore = new MongoDBAtlasVectorSearch({
mongodbClient: client,
dbName: "nerdWorkState",
collectionName: "nerdIndex", // this is where your embeddings will be stored
indexName: "nerd_index", // this is the name of the index you will need to create
});

// now create an index from all the Documents and store them in Atlas
const storageContext = await storageContextFromDefaults({ vectorStore });

const essay = payload;

// Create Document object with essay
const document = new Document({ text: essay, id_: id });
console.log({ document });
// Split text and create embeddings. Store them in a VectorStoreIndex
const result = await VectorStoreIndex.fromDocuments([document], { storageContext });
const embeddingResults = await result.getNodeEmbeddingResults([document]);
console.log({ result, embeddingResults });
const db = client.db("nerdWorkState"); // Connect to the database
const hackIndex = db.collection("nerdIndex");

const embedding = await hackIndex.findOne({ "metadata.doc_id": id });

console.log({ embeddingId: embedding?.id });
console.log(`Successfully created embeddings in the MongoDB collection`);
return { embeddingId: embedding?.id as string, result: embeddingResults };
}

async function runLlamaAndStore(
db: Db,
usedEmbeddingIds: string[],
promptMessages: any,
haikipu: Haikipu,
) {
const haikuId = haikipu.id;
const { embeddingId } = await llamaindex(JSON.stringify(haikipu), haikuId); //should we modify this id?
// store in DB
const promptResult = await storePrompt(
db,
haikipu,
promptMessages,
embeddingId,
usedEmbeddingIds,
);
const usageResult = await storeUsageByEmbeddingId(db, haikuId, embeddingId, usedEmbeddingIds);
const evaluationResult = await storeEvaluationByProject(db, haikuId, usedEmbeddingIds, embeddingId, haikipu);
return {
promptResult,
usageResult,
evaluationResult,
};
async function runLlamaAndStore(db: Db, usedEmbeddingIds: string[], promptMessages: any, haikipu: Haikipu) {
const haikuId = haikipu.id;
const { embeddingId } = await llamaindex(JSON.stringify(haikipu), haikuId); //should we modify this id?
// store in DB
const promptResult = await storePrompt(db, haikipu, promptMessages, embeddingId, usedEmbeddingIds);
const usageResult = await storeUsageByEmbeddingId(db, haikuId, embeddingId, usedEmbeddingIds);
const evaluationResult = await storeEvaluationByProject(db, haikuId, usedEmbeddingIds, embeddingId, haikipu);
return {
promptResult,
usageResult,
evaluationResult,
};
}

// Revised function suited for hackathon application data
async function generateHackathonProposal(haikiput: Haikipu, systemPrompt: string, assistantPrompt?: string, userPrompt?: string) {
const messages: ChatMessage[] = [
{
role: "system",
content: systemPrompt
},
{
role: "assistant",
content: assistantPrompt
},
{
role: "user",
content: userPrompt
},
];

const llm = new OpenAI({
model: (process.env.MODEL as any) ?? "gpt-4-0125-preview",
maxTokens: 512,
additionalChatOptions: { response_format: { type: "json_object" } },
});

const serviceContext = serviceContextFromDefaults({
llm,
chunkSize: 512,
chunkOverlap: 20,
});

const chatEngine = await createChatEngine(serviceContext);
if (!chatEngine) {
throw new Error("datasource is required in the request body");
}

// Convert message content from Vercel/AI format to LlamaIndex/OpenAI format

const response = await chatEngine.chat({
message: "Evaluate the summary and create a Haikipu.",
chatHistory: messages,
});
console.log({
response,
serviceContext,
raw: response.response,
firstNode: !!response.sourceNodes?.length && response.sourceNodes[0].hash,
});
const usedEmbeddingIds = response.sourceNodes?.map(node => node.id_) || [];

const parsedResponse = JSON.parse(response.response);

const haikipu: Haikipu = {
title: haikiput.title,
id: haikiput.id,
address: haikiput.address,
timestamp: Date.now().toString(),
type: haikiput.type,
contextSummary: haikiput.contextSummary,
haiku: parsedResponse.haiku,
explainer: parsedResponse.haikuExplainer,
};

return { haikipu, messages, usedEmbeddingIds };
async function generateHackathonProposal(
haikiput: Haikipu,
systemPrompt: string,
assistantPrompt?: string,
userPrompt?: string,
) {
const messages: ChatMessage[] = [
{
role: "system",
content: systemPrompt,
},
{
role: "assistant",
content: assistantPrompt,
},
{
role: "user",
content: userPrompt,
},
];

const llm = new OpenAI({
model: (process.env.MODEL as any) ?? "gpt-4-0125-preview",
maxTokens: 512,
additionalChatOptions: { response_format: { type: "json_object" } },
});

const serviceContext = serviceContextFromDefaults({
llm,
chunkSize: 512,
chunkOverlap: 20,
});

const chatEngine = await createChatEngine(serviceContext);
if (!chatEngine) {
throw new Error("datasource is required in the request body");
}

// Convert message content from Vercel/AI format to LlamaIndex/OpenAI format

const response = await chatEngine.chat({
message: "Evaluate the summary and create a Haikipu.",
chatHistory: messages,
});
console.log({
response,
serviceContext,
raw: response.response,
firstNode: !!response.sourceNodes?.length && response.sourceNodes[0].hash,
});
const usedEmbeddingIds = response.sourceNodes?.map(node => node.id_) || [];

const parsedResponse = JSON.parse(response.response);

const haikipu: Haikipu = {
title: haikiput.title,
id: haikiput.id,
address: haikiput.address,
timestamp: Date.now().toString(),
type: haikiput.type,
contextSummary: haikiput.contextSummary,
haiku: parsedResponse.haiku,
explainer: parsedResponse.explainer,
};

return { haikipu, messages, usedEmbeddingIds };
}

export const maxDuration = 120; // This function can run for a maximum of 5 seconds
// Example usage for POST handler or another part of your application
export async function hAIku(haikiput: Haikipu, systemPrompt: string, assistantPrompt?: string, userPrompt?: string) {
console.log(haikiput);
const { usedEmbeddingIds, messages, haikipu } = await generateHackathonProposal(
haikiput,
systemPrompt,
assistantPrompt,
userPrompt,
);
// Proceed with storing the enhanced proposal in MongoDB or returning it in the response
//
const db = client.db("nerdWorkState"); // Connect to the database
const haikuCodex = db.collection("nerdHaikus"); //
// assumed input
// run this function asynchronously, do not block for it to finish
runLlamaAndStore(db, usedEmbeddingIds, messages, haikipu);

await haikuCodex.updateOne(
{
id: haikipu.id,
address: haikipu.address,
haikipu,
},
{ $setOnInsert: { haikipu: haikipu } },
{ upsert: true }, // this creates new document if none match the filter
);
//
return haikipu;
console.log(haikiput);
const { usedEmbeddingIds, messages, haikipu } = await generateHackathonProposal(
haikiput,
systemPrompt,
assistantPrompt,
userPrompt,
);
// Proceed with storing the enhanced proposal in MongoDB or returning it in the response
//
const db = client.db("nerdWorkState"); // Connect to the database
const haikuCodex = db.collection("nerdHaikus"); //
// assumed input
// run this function asynchronously, do not block for it to finish
runLlamaAndStore(db, usedEmbeddingIds, messages, haikipu);

await haikuCodex.updateOne(
{
id: haikipu.id,
address: haikipu.address,
haikipu,
},
{ $setOnInsert: { haikipu: haikipu } },
{ upsert: true }, // this creates new document if none match the filter
);
//
return haikipu;
}

export const someVar = async (c: any, next: any) => {
const test = {} as Haikipu
const test = {} as Haikipu;

console.log(c.body)
const result = await hAIku(test, c.buttonValue, "", "")
console.log(c.body);
const result = await hAIku(test, c.buttonValue, "", "");

c._var.text = result.haiku
await next()
}
c._var.text = result.haiku;
await next();
};
2 changes: 1 addition & 1 deletion packages/nextjs/app/frames/tx-success/route.tsx
View file
Original file line number Diff line number Diff line change
Expand Up @@ -86,7 +86,7 @@ export const POST = frames(async ctx => {
await inngest.send({
name: "test/hello4.world",
data: {
prompt: data,
prompt: data[0],
fid: latestLog.args?.tokenId,
},
});
Expand Down
Loading

0 comments on commit 985f59c

Please sign in to comment.