Logging media

Weave supports logging and displaying multiple first class media types. Log images with PIL.Image.Image and audio with wave.Wave_read either directly with the object API, or as the inputs or output of an op.

Images

Logging type: PIL.Image.Image. Here is an example of logging an image with the OpenAI DALL-E API:

Python

import weave
from openai import OpenAI
import requests
from PIL import Image

weave.init('image-example')
client = OpenAI()

@weave.op
def generate_image(prompt: str) -> Image:
    response = client.images.generate(
        model="dall-e-3",
        prompt=prompt,
        size="1024x1024",
        quality="standard",
        n=1,
    )
    image_url = response.data[0].url
    image_response = requests.get(image_url, stream=True)
    image = Image.open(image_response.raw)

    # return a PIL.Image.Image object to be logged as an image
    return image

generate_image("a cat with a pumpkin hat")

TypeScript

import {OpenAI} from 'openai';
import * as weave from 'weave';

async function main() {
    const client = await weave.init('image-example');
    const openai = new OpenAI();

    const generateImage = weave.op(async (prompt: string) => {
        const response = await openai.images.generate({
            model: 'dall-e-3',
            prompt: prompt,
            size: '1024x1024',
            quality: 'standard',
            n: 1,
        });
        const imageUrl = response.data[0].url;
        const imgResponse = await fetch(imageUrl);
        const data = Buffer.from(await imgResponse.arrayBuffer());

        return weave.weaveImage({data});
    });

    generateImage('a cat with a pumpkin hat');
}

main();

This image will be logged to weave and automatically displayed in the UI. The following is the trace view for above.

Audio

Logging type: wave.Wave_read. Here is an example of logging an audio file using openai's speech generation API.

Python

import weave
from openai import OpenAI
import wave

weave.init("audio-example")
client = OpenAI()


@weave.op
def make_audio_file_streaming(text: str) -> wave.Wave_read:
    with client.audio.speech.with_streaming_response.create(
        model="tts-1",
        voice="alloy",
        input=text,
        response_format="wav",
    ) as res:
        res.stream_to_file("output.wav")

    # return a wave.Wave_read object to be logged as audio
    return wave.open("output.wav")

make_audio_file_streaming("Hello, how are you?")

TypeScript

import {OpenAI} from 'openai';
import * as weave from 'weave';

async function main() {
    await weave.init('audio-example');
    const openai = new OpenAI();

    const makeAudioFileStreaming = weave.op(async function audio(text: string) {
        const response = await openai.audio.speech.create({
            model: 'tts-1',
            voice: 'alloy',
            input: text,
            response_format: 'wav',
        });

        const chunks: Uint8Array[] = [];
        for await (const chunk of response.body) {
            chunks.push(chunk);
        }
        return weave.weaveAudio({data: Buffer.concat(chunks)});
    });

    await makeAudioFileStreaming('Hello, how are you?');
}

main();

This audio will be logged to weave and automatically displayed in the UI, with an audio player. The player can be expanded to view the raw audio waveform, in addition to a download button.

Try our cookbook for Audio Logging or

Open in Colab

. The cookbook also includes an advanced example of a Real Time Audio API based assistant integrated with Weave.