Request to stream a chat completion results in buffered response instead of stream

[au-re]

What Happened?

I tried running the gateway locally using npx @portkey-ai/gateway, node run dev:node and node build/start-server.js (Node 18). I noticed that even though I was specifying the "stream" parameter in the config the response was buffered and returned all at once to the client instead of streamed back. I also noticed that this is somehow related to the "compress" middleware step. After removing the following middleware, the response was streamed correctly:

// app.use('*', (c, next) => {
//   const runtime = getRuntimeKey();
//   if (runtime !== 'lagon' && runtime !== 'workerd') {
//     return compress()(c, next);
//   }
//   return next();
// });

What Should Have Happened?

The response should be streamed to the client.

Relevant Code Snippet

code on the client side

const run = async () => {

    const client = new OpenAI({
        apiKey: "xxx",
        baseURL: "http://127.0.0.1:8787/v1",
        defaultHeaders: {
            'x-portkey-config': JSON.stringify({
                "provider": "openai",
                "api_key": "xxx",
            })
        },
    });

    const stream = await client.chat.completions.create({
        model: "gpt-3.5-turbo",
        messages: [{ role: "system", content: "You are a helpful assistant." }, { role: "user", content: "Write a short story about Mumin" }],
        stream: true,
        max_tokens: 254,
    });

    for await (const part of stream) {
        console.log(part.choices[0]?.delta?.content || "", new Date());
    }
}

run()

Your Twitter/LinkedIn

No response

[mohankumarelec]

I am also facing same issue here. Removing

// index.ts
app.use('*', (c, next) => {
    const runtime = getRuntimeKey();
    if (runtime !== 'lagon' && runtime !== 'workerd') {
      return compress()(c, next);
    }
    return next();
});

and removing content-encoding by adding below in updateResponseHeaders

// handlerUtils.ts
response.headers.delete('content-encoding')

Solved it. Any permanent fix available ? @VisargD