HuggingFaceStream is giving stream not compatible with StreamingTextResponse

[Patrick McEnaney]

Description

Previously I was using AI SDK v2.2.37. which was working fine with huggingface untill I had to implement Google Generative AI. I did not wanted to implement legacy Google Generative AI so I updated ai SDK to v3.2.37.

after update, GoogleGenerative AI is working as well as legacy huggingface locally, but when deployed on vercel preview, huggingface is giving error when stream generated from HuggingFaceStream passed to StreamingTextResponse getting 405 Method not Allowed.
Same code works for google and huggingface locally.

Is this bug in package or in vercel?

Code example

import { authOptions } from "@/auth";
import { getServerSession } from "next-auth";
import { HuggingFaceStream, StreamingTextResponse } from "ai";

import { getModelByName } from "@/app/lib/models";
import {
  getGoogleChatResponseAsStream,
  getHuggingfaceChatResponseAsStream,
} from "@/app/lib/ai-providers";
import { textGenerationStream } from "@huggingface/inference";
import { getStreamEvents } from "@/app/lib/ai-providers/stream-events";

export async function POST(req: Request) {
  const token = process.env.HUGGINGFACE_API_KEY;
  const session = await getServerSession(authOptions);

  if (!session) {
    return new Response(null, { status: 401, statusText: "unauthorized" });
  }

  const userId = session?.user.id;

  const {
    messages,
    id: chatId,
    aiModel,
    systemPrompt,
    temperature,
    topK,
    topP,
    repetitionPenalty,
  } = await req.json();

  const lastMsg = messages[messages.length - 1];

  try {
    const modelSpec = getModelByName(aiModel);
    if (!modelSpec) {
      return new Response("Model not found", { status: 400, statusText: "Model not found" });
    }

    const parameters = {
      ...modelSpec.options.parameters,
      systemPrompt:
        systemPrompt !== undefined ? systemPrompt : modelSpec.options.parameters.systemPrompt,
      temperature:
        temperature !== undefined
          ? parseFloat(String(temperature))
          : modelSpec.options.parameters.temperature,
      top_k: topK !== undefined ? parseInt(String(topK)) : modelSpec.options.parameters.top_k,
      top_p: topP !== undefined ? parseFloat(String(topP)) : modelSpec.options.parameters.top_p,
      repetition_penalty:
        repetitionPenalty !== undefined
          ? parseFloat(String(repetitionPenalty))
          : modelSpec.options.parameters.repetition_penalty,
    };
    const prompt = modelSpec.promptConstructor([{ role: "user", content: lastMsg.content }]);

    let stream: ReadableStream;
    if (modelSpec.provider === "huggingface") {
      // stream = getHuggingfaceChatResponseAsStream(prompt, {
      //   userId,
      //   lastMsg,
      //   chatId,
      //   aiModel,
      //   parameters,
      //   token,
      //   otherOptions: modelSpec.options,
      // });
      const response = textGenerationStream({
        inputs: prompt,
        accessToken: token,
        ...modelSpec.options,
        parameters,
      });
      stream = HuggingFaceStream(
        response,
        getStreamEvents(userId, lastMsg, chatId, aiModel, parameters),
      );
    } else if (modelSpec.provider === "google") {
      stream = await getGoogleChatResponseAsStream(prompt, {
        userId,
        lastMsg,
        chatId,
        aiModel,
        parameters,
      });
    } else {
      stream = new ReadableStream();
    }

    return new StreamingTextResponse(stream);
  } catch (error) {
    return new Response("Service unavailable", { status: 500, statusText: JSON.stringify(error) });
  }
}

Additional context

getStreamEvents() is simply returning and object with onStart and onFinish callbacks.

{
    onStart:()=>{},
    onFinish:(completition:string) => {},
}