Description
It appears that the BatchRequestOutput object's error is missing properties and in an error response, that the actual error contents are nested under the body property.
The spec shows that a BatchRequestOutput object will contain id, custom_id, response, and error, and that the error property will have code and message:
However, when I construct a request that's intentionally designed to trigger a non-HTTP error (set max_tokens to 1000000 for a gpt-4o prompt), create/upload a file containing this request via the files endpoint, trigger a batch request with this file, and then retrieve the contents of the resulting Error file to inspect its contents, I see the following discrepancies:
- the
errorproperty is nulled, and theresponse.bodycontains the actual error contents - the
errorobject not only has thecodeandmessageproperties, but also hastypeandparamproperties
request:
{
"custom_id": "request-5",
"method": "POST",
"url": "/v1/chat/completions",
"body": {
"model": "gpt-4o",
"messages": [
{
"role": "system",
"content": "You are a helpful assistant."
},
{
"role": "user",
"content": "What is the capital of Texas?"
}
],
"max_tokens": 1000000
}
}
response:
"id": "batch_req_YTpodqz4ZJ8arId5Bm73lIEt",
"custom_id": "request-5",
"response": {
"status_code": 400,
"request_id": "0538a1cc8f8f8500bbe07250e6aa2b26",
"body": {
"error": {
"message": "This model's maximum context length is 128000 tokens. However, you requested 1000024 tokens (24 in the messages, 1000000 in the completion). Please reduce the length of the messages or completion.",
"type": "invalid_request_error",
"param": "messages",
"code": "context_length_exceeded"
}
}
},
"error": null
}
It appears that the spec should at least add the type and param properties to error, and that error should be 'moved'?