Open
Description
Description of the bug:
If you submit a prompt like this to Gemini Pro:
{
contents: [
{ role: "user", ...}
{ role: "user", ....}
]
}
The API will respond that the last message must come from user.
This occurred when I was translating my GPT role: "system" messages to be role: "user". (But GPT also doesn't reject multiple user messages in a row)
Actual vs expected behavior:
- The error message should be accurate
- The API shouldn't reject two user messages in a row
Any other information you'd like to share?
No response