Openai batch api polling alternative

API
1

Currently when I send off my data to be scored by openai, I continuously poll to check if the results are back. This is fine if the data returns quickly, but given the 24hr turn around for the batch api, this can take a while.

Wondering if there is a work around for this. eg. send me an email when the processing is done and I can download results from a openai link. My code is shown below:

    for chunk_idx, df_chunk in enumerate(np.array_split(df, num_chunks)):
        requests, requests_customids = create_batch_requests(
            df_chunk, gpt4o_version, temperature, image_detail
        )
        save_jsonl(requests, f"batch_input_chunk_{chunk_idx + 1}.jsonl")
        batch_job_id = run_batch_job(openai, f"batch_input_chunk_{chunk_idx + 1}.jsonl")

        while True:
            batch_status = check_batch_status(openai, batch_job_id)
            if batch_status.output_file_id:
                output_file_id = batch_status.output_file_id
                break
            sleep(60)  # check every minute

        batch_data = extract_data(openai, output_file_id)
        df_batch_output = pd.DataFrame(batch_data)

where:

def run_batch_job(client: openai, input_file: str) -> str:
    batch_input_file = client.files.create(file=open(input_file, "rb"), purpose="batch")
    batch_input_file_id = batch_input_file.id

    # create batch job (this step costs money!!!)
    batch_job = client.batches.create(
        input_file_id=batch_input_file_id, endpoint="/v1/chat/completions", completion_window="24h"
    )

    return batch_job.id


def check_batch_status(client: openai, job_id: str):
    # check the status of batch job to see if it's completed
    retrieve_batch = client.batches.retrieve(job_id)
    return retrieve_batch
2

No. But your polling request response contains the number of completed rows, so you can estimate completion time.