Rayy•11mo ago

Is there a way to run mutations in bulk?

I am trying to create embeddings from chunk and then store those embedding in the convex vector store. But rather than calling one mutation for each embedding response, is there a way I can add them in bulk? I am trying to minimize addEmbedding calls.

export const generateEmbeddings = internalAction({
  args: {...},
  handler: async (ctx, args) => {
    //Batching chunks

    const responses = await Promise.all(
      batches.map((batch) => fetchEmbedding(batch))
    );

    responses.forEach(async (response) => {
      for (let i = 0; i < response.data.length; i++) {
        const chunk =
          typeof args.chunks[0] !== "string"
            ? (args.chunks[i] as Chunk)
            : undefined;
        await ctx.runMutation(internal.chatbook.addEmbedding, {
          chatId: args.chatId,
          content: chunk?.content ?? (args.chunks[i] as string),
          source: chunk?.source ?? "",
          embedding: response.data[i].embedding,
          title: args.title,
          type: args.type,
        });
      }
    });
  },
});

export const addEmbedding = internalMutation({
  args: {
    chatId: v.id("chatbook"),
    content: v.string(),
    source: v.optional(v.string()),
    embedding: v.array(v.number()),
    title: v.string(),
    type: v.union(v.literal("code"), v.literal("doc"), v.literal("video")),
  },
  handler: async (ctx, args) => {
    const chatEmbeddingId = await ctx.db.insert("chatEmbeddings", {
      embedding: args.embedding,
      content: args.content,
      chatId: args.chatId,
      source: args.source,
    });
    const existingChatDocument = await ctx.db.get(args.chatId);
    const embeddingIds = existingChatDocument?.embeddingId || [];

    embeddingIds.push(chatEmbeddingId);

    await ctx.db.patch(args.chatId, {
      embeddingId: embeddingIds,
      title: args.title,
      type: args.type,
    });
  },
});

export const generateEmbeddings = internalAction({
  args: {...},
  handler: async (ctx, args) => {
    //Batching chunks

    const responses = await Promise.all(
      batches.map((batch) => fetchEmbedding(batch))
    );

    responses.forEach(async (response) => {
      for (let i = 0; i < response.data.length; i++) {
        const chunk =
          typeof args.chunks[0] !== "string"
            ? (args.chunks[i] as Chunk)
            : undefined;
        await ctx.runMutation(internal.chatbook.addEmbedding, {
          chatId: args.chatId,
          content: chunk?.content ?? (args.chunks[i] as string),
          source: chunk?.source ?? "",
          embedding: response.data[i].embedding,
          title: args.title,
          type: args.type,
        });
      }
    });
  },
});

export const addEmbedding = internalMutation({
  args: {
    chatId: v.id("chatbook"),
    content: v.string(),
    source: v.optional(v.string()),
    embedding: v.array(v.number()),
    title: v.string(),
    type: v.union(v.literal("code"), v.literal("doc"), v.literal("video")),
  },
  handler: async (ctx, args) => {
    const chatEmbeddingId = await ctx.db.insert("chatEmbeddings", {
      embedding: args.embedding,
      content: args.content,
      chatId: args.chatId,
      source: args.source,
    });
    const existingChatDocument = await ctx.db.get(args.chatId);
    const embeddingIds = existingChatDocument?.embeddingId || [];

    embeddingIds.push(chatEmbeddingId);

    await ctx.db.patch(args.chatId, {
      embeddingId: embeddingIds,
      title: args.title,
      type: args.type,
    });
  },
});

2 Replies

ian•11mo ago

Just pass the array in and pass an array back- it’s all just code. Nothing requires you to only take one parameter. Maybe call it addEmbeddingBatch But don’t pass all at once- you should pass them in chunks, maybe start with 100 and see?

RayyOP•11mo ago

Ah yes, ofc. That was such a simple thing. Thank you so much. 🙂

Is there a way to run mutations in bulk?

Did you find this page helpful?