bug(node, vectordb): error when using Float64 embedding function

[universalmind303]

LanceDB version

main

What happened?

I get some buffer allocation errors if i try to specify Float64 as my embedding function type

Error: Invalid argument error: Need at least 8192 bytes in buffers[0] in array of type Float64, but got 4096

Are there known steps to reproduce?

const db = await lancedb.connect("/tmp/lancedb");

const embeddingFunction = {
  sourceColumn: "text",
  embed: async function (batch) {
    return batch.map(() => Array(512).fill(1.0))

  },
};

const data = [
  {text: "hello"},
  {text: "hello world"},
]

const schema = new Schema([
  new Field("vector", new FixedSizeList(512, new Field("item", new Float64()))),
  new Field("text", new Utf8())
]);

const name = "test";
let table = await db.createTable({name, data, schema, embeddingFunction})

However, if I change Float64 to Float32, everything works fine.

[wjones127]

That seems odd. We might hard code f32 in some places, but in Lance we do support f16 and f64, so we should fix any places that assume f32.