Support for the promptFeedback.blockReason in the Gemini streaming output

[xukecheng]

Description

During the use of Gemini API, it is possible to encounter a return blockReason.

{
  "promptFeedback": {
    "blockReason": "PROHIBITED_CONTENT"
  },
  "usageMetadata": {
    "promptTokenCount": 12559,
    "totalTokenCount": 12559
  },
  "modelVersion": "gemini-1.5-flash-002"
}

However, it currently seems that there is no way to obtain the blockReason in the streaming output.

Code example

  const textStream = await streamText({
    model,
    system,
    prompt,
    maxTokens: 8192,
    onFinish({ usage }) {
      modelUsage = usage;
    },
  });

  for await (const textPart of textStream.textStream) {
    const formattedText = textPart.replace(/\n/g, '\r\n');
    controller.enqueue(encoder.encode(`data: ${formattedText}\n\n`));
  }

AI provider

[email protected] @ai-sdk/[email protected]

Additional context

No response