Redigeeri

Jagamisviis:

Getting started with Azure OpenAI global batch deployments

  • Artikkel

The Azure OpenAI Batch API is designed to handle large-scale and high-volume processing tasks efficiently. Process asynchronous groups of requests with separate quota, with 24-hour target turnaround, at 50% less cost than global standard. With batch processing, rather than send one request at a time you send a large number of requests in a single file. Global batch requests have a separate enqueued token quota avoiding any disruption of your online workloads.

Key use cases include:

  • Large-Scale Data Processing: Quickly analyze extensive datasets in parallel.

  • Content Generation: Create large volumes of text, such as product descriptions or articles.

  • Document Review and Summarization: Automate the review and summarization of lengthy documents.

  • Customer Support Automation: Handle numerous queries simultaneously for faster responses.

  • Data Extraction and Analysis: Extract and analyze information from vast amounts of unstructured data.

  • Natural Language Processing (NLP) Tasks: Perform tasks like sentiment analysis or translation on large datasets.

  • Marketing and Personalization: Generate personalized content and recommendations at scale.

Important

We aim to process batch requests within 24 hours; we do not expire the jobs that take longer. You can cancel the job anytime. When you cancel the job, any remaining work is cancelled and any already completed work is returned. You will be charged for any completed work.

Data stored at rest remains in the designated Azure geography, while data may be processed for inferencing in any Azure OpenAI location. Learn more about data residency. 

Global batch support

Region and model support

Global batch is currently supported in the following regions:

Region gpt-4o, 2024-05-13 gpt-4o, 2024-08-06 gpt-4o-mini, 2024-07-18 gpt-4, 0613 gpt-4, turbo-2024-04-09 gpt-35-turbo, 0613 gpt-35-turbo, 1106 gpt-35-turbo, 0125
canadaeast ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
eastus ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
eastus2 ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
northcentralus ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
southcentralus ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
swedencentral ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
westus ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅
westus3 ✅ ✅ ✅ ✅ ✅ ✅ ✅ ✅

The following models support global batch:

Model Version Input format
gpt-4o 2024-08-06 text + image
gpt-4o-mini 2024-07-18 text + image
gpt-4o 2024-05-13 text + image
gpt-4 turbo-2024-04-09 text
gpt-4 0613 text
gpt-35-turbo 0125 text
gpt-35-turbo 1106 text
gpt-35-turbo 0613 text

Refer to the models page for the most up-to-date information on regions/models where global batch is currently supported.

API support

API Version
Latest GA API release: 2024-10-21
Latest Preview API release: 2024-10-01-preview

Support first added in: 2024-07-01-preview

Feature support

The following aren't currently supported:

  • Integration with the Assistants API.
  • Integration with Azure OpenAI On Your Data feature.

Note

Structured outputs is now supported with Global Batch.

Global batch deployment

In the Studio UI the deployment type will appear as Global-Batch.

Screenshot that shows the model deployment dialog in Azure AI Studio with Global-Batch deployment type highlighted.

Tip

We recommend enabling dynamic quota for all global batch model deployments to help avoid job failures due to insufficient enqueued token quota. Dynamic quota allows your deployment to opportunistically take advantage of more quota when extra capacity is available. When dynamic quota is set to off, your deployment will only be able to process requests up to the enqueued token limit that was defined when you created the deployment.

Prerequisites

Preparing your batch file

Like fine-tuning, global batch uses files in JSON lines (.jsonl) format. Below are some example files with different types of supported content:

Input format

{"custom_id": "task-0", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "When was Microsoft founded?"}]}}
{"custom_id": "task-1", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "When was the first XBOX released?"}]}}
{"custom_id": "task-2", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "What is Altair Basic?"}]}}

The custom_id is required to allow you to identify which individual batch request corresponds to a given response. Responses won't be returned in identical order to the order defined in the .jsonl batch file.

model attribute should be set to match the name of the Global Batch deployment you wish to target for inference responses.

Important

The model attribute must be set to match the name of the Global Batch deployment you wish to target for inference responses. The same Global Batch model deployment name must be present on each line of the batch file. If you want to target a different deployment you must do so in a separate batch file/job.

For the best performance we recommend submitting large files for batch processing, rather than a large number of small files with only a few lines in each file.

Create input file

For this article, we'll create a file named test.jsonl and will copy the contents from standard input code block above to the file. You'll need to modify and add your global batch deployment name to each line of the file.

Upload batch file

Once your input file is prepared, you first need to upload the file to then be able to kick off a batch job. File upload can be done both programmatically or via the Studio.

  1. Sign in to AI Studio.

  2. Select the Azure OpenAI resource where you have a global batch model deployment available.

  3. Select Batch jobs > +Create batch jobs.

    Screenshot that shows the batch job creation experience in Azure AI Studio.

  4. From the dropdown under Batch data > Upload files > select Upload file and provide the path for the test.jsonl file created in the previous step > Next.

    Screenshot that shows upload file experience.

Create batch job

Select Create to start your batch job.

Screenshot of the create a batch job Azure Studio UI experience.

Track batch job progress

Once your job is created, you can monitor the job's progress by selecting the Job ID for the most recently created job. By default you will be taken to the status page for your most recently created batch job.

Screenshot that shows the batch job ID for a job currently undergoing validation.

You can track job status for your job in the right-hand pane:

Screenshot that shows the batch job status experience in Azure AI Studio.

Retrieve batch job output file

Once your job has completed or reached a terminal state, it will generate an error file and an output file which can be downloaded for review by selecting the respective button with the downward arrow icon.

Screenshot that shows the batch job output and error files available for download.

Cancel batch

Cancels an in-progress batch. The batch will be in status cancelling for up to 10 minutes, before changing to cancelled, where it will have partial results (if any) available in the output file.

Screenshot that shows the batch job cancel button in Azure AI Studio.

Prerequisites

The steps in this article are intended to be run sequentially in Jupyter Notebooks. For this reason we'll only instantiate the Azure OpenAI client once at the beginning of the examples. If you want to run a step out-of-order you'll often need to set up an Azure OpenAI client as part of that call.

Even if you already have the OpenAI Python library installed you might need to upgrade your installation to the latest version:

!pip install openai --upgrade

Preparing your batch file

Like fine-tuning, global batch uses files in JSON lines (.jsonl) format. Below are some example files with different types of supported content:

Input format

{"custom_id": "task-0", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "When was Microsoft founded?"}]}}
{"custom_id": "task-1", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "When was the first XBOX released?"}]}}
{"custom_id": "task-2", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "What is Altair Basic?"}]}}

The custom_id is required to allow you to identify which individual batch request corresponds to a given response. Responses won't be returned in identical order to the order defined in the .jsonl batch file.

model attribute should be set to match the name of the Global Batch deployment you wish to target for inference responses.

Important

The model attribute must be set to match the name of the Global Batch deployment you wish to target for inference responses. The same Global Batch model deployment name must be present on each line of the batch file. If you want to target a different deployment you must do so in a separate batch file/job.

For the best performance we recommend submitting large files for batch processing, rather than a large number of small files with only a few lines in each file.

Create input file

For this article we'll create a file named test.jsonl and will copy the contents from standard input code block above to the file. You will need to modify and add your global batch deployment name to each line of the file. Save this file in the same directory that you're executing your Jupyter Notebook.

Upload batch file

Once your input file is prepared, you first need to upload the file to then be able to kick off a batch job. File upload can be done both programmatically or via the Studio. This example uses environment variables in place of the key and endpoint values. If you're unfamiliar with using environment variables with Python refer to one of our quickstarts where the process of setting up the environment variables in explained step-by-step.

import os
from openai import AzureOpenAI
from azure.identity import DefaultAzureCredential, get_bearer_token_provider

token_provider = get_bearer_token_provider(
    DefaultAzureCredential(), "https://cognitiveservices.azure.com/.default"
)

client = AzureOpenAI(
  azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT"), 
  azure_ad_token_provider=token_provider,
  api_version="2024-10-21"
)

# Upload a file with a purpose of "batch"
file = client.files.create(
  file=open("test.jsonl", "rb"), 
  purpose="batch"
)

print(file.model_dump_json(indent=2))
file_id = file.id

Output:

{
  "id": "file-9f3a81d899b4442f98b640e4bc3535dd",
  "bytes": 815,
  "created_at": 1722476551,
  "filename": "test.jsonl",
  "object": "file",
  "purpose": "batch",
  "status": null,
  "status_details": null
}

Create batch job

Once your file has uploaded successfully you can submit the file for batch processing.

# Submit a batch job with the file
batch_response = client.batches.create(
    input_file_id=file_id,
    endpoint="/chat/completions",
    completion_window="24h",
)

# Save batch ID for later use
batch_id = batch_response.id

print(batch_response.model_dump_json(indent=2))

Note

Currently the completion window must be set to 24h. If you set any other value than 24h your job will fail. Jobs taking longer than 24 hours will continue to execute until cancelled.

Output:

{
  "id": "batch_6caaf24d-54a5-46be-b1b7-518884fcbdde",
  "completion_window": "24h",
  "created_at": 1722476583,
  "endpoint": null,
  "input_file_id": "file-9f3a81d899b4442f98b640e4bc3535dd",
  "object": "batch",
  "status": "validating",
  "cancelled_at": null,
  "cancelling_at": null,
  "completed_at": null,
  "error_file_id": null,
  "errors": null,
  "expired_at": null,
  "expires_at": 1722562983,
  "failed_at": null,
  "finalizing_at": null,
  "in_progress_at": null,
  "metadata": null,
  "output_file_id": null,
  "request_counts": {
    "completed": 0,
    "failed": 0,
    "total": 0
  }
}

Track batch job progress

Once you have created batch job successfully you can monitor its progress either in the Studio or programatically. When checking batch job progress we recommend waiting at least 60 seconds in between each status call.

import time
import datetime 

status = "validating"
while status not in ("completed", "failed", "canceled"):
    time.sleep(60)
    batch_response = client.batches.retrieve(batch_id)
    status = batch_response.status
    print(f"{datetime.datetime.now()} Batch Id: {batch_id},  Status: {status}")

if batch_response.status == "failed":
    for error in batch_response.errors.data:  
        print(f"Error code {error.code} Message {error.message}")

Output:

2024-07-31 21:48:32.556488 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: validating
2024-07-31 21:49:39.221560 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: in_progress
2024-07-31 21:50:53.383138 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: in_progress
2024-07-31 21:52:07.274570 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: in_progress
2024-07-31 21:53:21.149501 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: finalizing
2024-07-31 21:54:34.572508 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: finalizing
2024-07-31 21:55:35.304713 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: finalizing
2024-07-31 21:56:36.531816 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: finalizing
2024-07-31 21:57:37.414105 Batch Id: batch_6caaf24d-54a5-46be-b1b7-518884fcbdde,  Status: completed

The following status values are possible:

Status Description
validating The input file is being validated before the batch processing can begin.
failed The input file has failed the validation process.
in_progress The input file was successfully validated and the batch is currently running.
finalizing The batch has completed and the results are being prepared.
completed The batch has been completed and the results are ready.
expired The batch wasn't able to be completed within the 24-hour time window.
cancelling The batch is being cancelled (This may take up to 10 minutes to go into effect.)
cancelled the batch was cancelled.

To examine the job status details you can run:

print(batch_response.model_dump_json(indent=2))

Output:

{
  "id": "batch_6caaf24d-54a5-46be-b1b7-518884fcbdde",
  "completion_window": "24h",
  "created_at": 1722476583,
  "endpoint": null,
  "input_file_id": "file-9f3a81d899b4442f98b640e4bc3535dd",
  "object": "batch",
  "status": "completed",
  "cancelled_at": null,
  "cancelling_at": null,
  "completed_at": 1722477429,
  "error_file_id": "file-c795ae52-3ba7-417d-86ec-07eebca57d0b",
  "errors": null,
  "expired_at": null,
  "expires_at": 1722562983,
  "failed_at": null,
  "finalizing_at": 1722477177,
  "in_progress_at": null,
  "metadata": null,
  "output_file_id": "file-3304e310-3b39-4e34-9f1c-e1c1504b2b2a",
  "request_counts": {
    "completed": 3,
    "failed": 0,
    "total": 3
  }
}

Observe that there's both error_file_id and a separate output_file_id. Use the error_file_id to assist in debugging any issues that occur with your batch job.

Retrieve batch job output file

import json

output_file_id = batch_response.output_file_id

if not output_file_id:
    output_file_id = batch_response.error_file_id

if output_file_id:
    file_response = client.files.content(output_file_id)
    raw_responses = file_response.text.strip().split('\n')  

    for raw_response in raw_responses:  
        json_response = json.loads(raw_response)  
        formatted_json = json.dumps(json_response, indent=2)  
        print(formatted_json)

Output:

For brevity, we are only including a single chat completion response of output. If you follow the steps in this article you should have three responses similar to the one below:

{
  "custom_id": "task-0",
  "response": {
    "body": {
      "choices": [
        {
          "content_filter_results": {
            "hate": {
              "filtered": false,
              "severity": "safe"
            },
            "self_harm": {
              "filtered": false,
              "severity": "safe"
            },
            "sexual": {
              "filtered": false,
              "severity": "safe"
            },
            "violence": {
              "filtered": false,
              "severity": "safe"
            }
          },
          "finish_reason": "stop",
          "index": 0,
          "logprobs": null,
          "message": {
            "content": "Microsoft was founded on April 4, 1975, by Bill Gates and Paul Allen in Albuquerque, New Mexico.",
            "role": "assistant"
          }
        }
      ],
      "created": 1722477079,
      "id": "chatcmpl-9rFGJ9dh08Tw9WRKqaEHwrkqRa4DJ",
      "model": "gpt-4o-2024-05-13",
      "object": "chat.completion",
      "prompt_filter_results": [
        {
          "prompt_index": 0,
          "content_filter_results": {
            "hate": {
              "filtered": false,
              "severity": "safe"
            },
            "jailbreak": {
              "filtered": false,
              "detected": false
            },
            "self_harm": {
              "filtered": false,
              "severity": "safe"
            },
            "sexual": {
              "filtered": false,
              "severity": "safe"
            },
            "violence": {
              "filtered": false,
              "severity": "safe"
            }
          }
        }
      ],
      "system_fingerprint": "fp_a9bfe9d51d",
      "usage": {
        "completion_tokens": 24,
        "prompt_tokens": 27,
        "total_tokens": 51
      }
    },
    "request_id": "660b7424-b648-4b67-addc-862ba067d442",
    "status_code": 200
  },
  "error": null
}

Additional batch commands

Cancel batch

Cancels an in-progress batch. The batch will be in status cancelling for up to 10 minutes, before changing to cancelled, where it will have partial results (if any) available in the output file.

client.batches.cancel("batch_abc123") # set to your batch_id for the job you want to cancel

List batch

List batch jobs for a particular Azure OpenAI resource.

client.batches.list()

List methods in the Python library are paginated.

To list all jobs:

all_jobs = []
# Automatically fetches more pages as needed.
for job in client.batches.list(
    limit=20,
):
    # Do something with job here
    all_jobs.append(job)
print(all_jobs)

List batch (Preview)

Use the REST API to list all batch jobs with additional sorting/filtering options.

In the examples below we are providing the generate_time_filter function to make constructing the filter easier. If you don't wish to use this function the format of the filter string would look like created_at gt 1728860560 and status eq 'Completed'.

import requests
import json
from datetime import datetime, timedelta
from azure.identity import DefaultAzureCredential

token_credential = DefaultAzureCredential()
token = token_credential.get_token('https://cognitiveservices.azure.com/.default')

endpoint = "https://{YOUR_RESOURCE_NAME}.openai.azure.com/"
api_version = "2024-10-01-preview"
url = f"{endpoint}openai/batches"
order = "created_at asc"
time_filter =  lambda: generate_time_filter("past 8 hours")

# Additional filter examples:
#time_filter =  lambda: generate_time_filter("past 1 day")
#time_filter =  lambda: generate_time_filter("past 3 days", status="Completed")

def generate_time_filter(time_range, status=None):
    now = datetime.now()
    
    if 'day' in time_range:
        days = int(time_range.split()[1])
        start_time = now - timedelta(days=days)
    elif 'hour' in time_range:
        hours = int(time_range.split()[1])
        start_time = now - timedelta(hours=hours)
    else:
        raise ValueError("Invalid time range format. Use 'past X day(s)' or 'past X hour(s)'")
    
    start_timestamp = int(start_time.timestamp())
    
    filter_string = f"created_at gt {start_timestamp}"
    
    if status:
        filter_string += f" and status eq '{status}'"
    
    return filter_string

filter = time_filter()

headers = {'Authorization': 'Bearer ' + token.token}

params = {
    "api-version": api_version,
    "$filter": filter,
    "$orderby": order
}

response = requests.get(url, headers=headers, params=params)

json_data = response.json()

if response.status_code == 200:
    print(json.dumps(json_data, indent=2))
else:
    print(f"Request failed with status code: {response.status_code}")
    print(response.text)

Output:

{
  "data": [
    {
      "cancelled_at": null,
      "cancelling_at": null,
      "completed_at": 1729011896,
      "completion_window": "24h",
      "created_at": 1729011128,
      "error_file_id": "file-472c0626-4561-4327-9e4e-f41afbfb30e6",
      "expired_at": null,
      "expires_at": 1729097528,
      "failed_at": null,
      "finalizing_at": 1729011805,
      "id": "batch_4ddc7b60-19a9-419b-8b93-b9a3274b33b5",
      "in_progress_at": 1729011493,
      "input_file_id": "file-f89384af0082485da43cb26b49dc25ce",
      "errors": null,
      "metadata": null,
      "object": "batch",
      "output_file_id": "file-62bebde8-e767-4cd3-a0a1-28b214dc8974",
      "request_counts": {
        "total": 3,
        "completed": 2,
        "failed": 1
      },
      "status": "completed",
      "endpoint": "/chat/completions"
    },
    {
      "cancelled_at": null,
      "cancelling_at": null,
      "completed_at": 1729016366,
      "completion_window": "24h",
      "created_at": 1729015829,
      "error_file_id": "file-85ae1971-9957-4511-9eb4-4cc9f708b904",
      "expired_at": null,
      "expires_at": 1729102229,
      "failed_at": null,
      "finalizing_at": 1729016272,
      "id": "batch_6287485f-50fc-4efa-bcc5-b86690037f43",
      "in_progress_at": 1729016126,
      "input_file_id": "file-686746fcb6bc47f495250191ffa8a28e",
      "errors": null,
      "metadata": null,
      "object": "batch",
      "output_file_id": "file-04399828-ae0b-4825-9b49-8976778918cb",
      "request_counts": {
        "total": 3,
        "completed": 2,
        "failed": 1
      },
      "status": "completed",
      "endpoint": "/chat/completions"
    }
  ],
  "first_id": "batch_4ddc7b60-19a9-419b-8b93-b9a3274b33b5",
  "has_more": false,
  "last_id": "batch_6287485f-50fc-4efa-bcc5-b86690037f43"
}

Prerequisites

Preparing your batch file

Like fine-tuning, global batch uses files in JSON lines (.jsonl) format. Below are some example files with different types of supported content:

Input format

{"custom_id": "task-0", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "When was Microsoft founded?"}]}}
{"custom_id": "task-1", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "When was the first XBOX released?"}]}}
{"custom_id": "task-2", "method": "POST", "url": "/chat/completions", "body": {"model": "REPLACE-WITH-MODEL-DEPLOYMENT-NAME", "messages": [{"role": "system", "content": "You are an AI assistant that helps people find information."}, {"role": "user", "content": "What is Altair Basic?"}]}}

The custom_id is required to allow you to identify which individual batch request corresponds to a given response. Responses won't be returned in identical order to the order defined in the .jsonl batch file.

model attribute should be set to match the name of the Global Batch deployment you wish to target for inference responses.

Important

The model attribute must be set to match the name of the Global Batch deployment you wish to target for inference responses. The same Global Batch model deployment name must be present on each line of the batch file. If you want to target a different deployment you must do so in a separate batch file/job.

For the best performance we recommend submitting large files for batch processing, rather than a large number of small files with only a few lines in each file.

Create input file

For this article we'll create a file named test.jsonl and will copy the contents from standard input code block above to the file. You will need to modify and add your global batch deployment name to each line of the file.

Upload batch file

Once your input file is prepared, you first need to upload the file to then be able to kick off a batch job. File upload can be done both programmatically or via the Studio. This example uses environment variables in place of the key and endpoint values. If you're unfamiliar with using environment variables with Python refer to one of our quickstarts where the process of setting up the environment variables in explained step-by-step.

Important

If you use an API key, store it securely somewhere else, such as in Azure Key Vault. Don't include the API key directly in your code, and never post it publicly.

For more information about AI services security, see Authenticate requests to Azure AI services.

curl -X POST https://YOUR_RESOURCE_NAME.openai.azure.com/openai/files?api-version=2024-10-21 \
  -H "Content-Type: multipart/form-data" \
  -H "api-key: $AZURE_OPENAI_API_KEY" \
  -F "purpose=batch" \
  -F "file=@C:\\batch\\test.jsonl;type=application/json"

The above code assumes a particular file path for your test.jsonl file. Adjust this file path as necessary for your local system.

Output:

{
  "status": "pending",
  "bytes": 686,
  "purpose": "batch",
  "filename": "test.jsonl",
  "id": "file-21006e70789246658b86a1fc205899a4",
  "created_at": 1721408291,
  "object": "file"
}

Track file upload status

Depending on the size of your upload file it might take some time before it's fully uploaded and processed. To check on your file upload status run:

curl https://YOUR_RESOURCE_NAME.openai.azure.com/openai/files/{file-id}?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY"

Output:

{
  "status": "processed",
  "bytes": 686,
  "purpose": "batch",
  "filename": "test.jsonl",
  "id": "file-21006e70789246658b86a1fc205899a4",
  "created_at": 1721408291,
  "object": "file"
}

Create batch job

Once your file has uploaded successfully you can submit the file for batch processing.

curl -X POST https://YOUR_RESOURCE_NAME.openai.azure.com/openai/batches?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "input_file_id": "file-abc123",
    "endpoint": "/chat/completions",
    "completion_window": "24h"
  }'

Note

Currently the completion window must be set to 24h. If you set any other value than 24h your job will fail. Jobs taking longer than 24 hours will continue to execute until cancelled.

Output:

{
  "cancelled_at": null,
  "cancelling_at": null,
  "completed_at": null,
  "completion_window": "24h",
  "created_at": "2024-07-19T17:13:57.2491382+00:00",
  "error_file_id": null,
  "expired_at": null,
  "expires_at": "2024-07-20T17:13:57.1918498+00:00",
  "failed_at": null,
  "finalizing_at": null,
  "id": "batch_fe3f047a-de39-4068-9008-346795bfc1db",
  "in_progress_at": null,
  "input_file_id": "file-21006e70789246658b86a1fc205899a4",
  "errors": null,
  "metadata": null,
  "object": "batch",
  "output_file_id": null,
  "request_counts": {
    "total": null,
    "completed": null,
    "failed": null
  },
  "status": "Validating"
}

Track batch job progress

Once you have created batch job successfully you can monitor its progress either in the Studio or programatically. When checking batch job progress we recommend waiting at least 60 seconds in between each status call.

curl https://YOUR_RESOURCE_NAME.openai.azure.com/openai/batches/{batch_id}?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY"

Output:

{
  "cancelled_at": null,
  "cancelling_at": null,
  "completed_at": null,
  "completion_window": "24h",
  "created_at": "2024-07-19T17:33:29.1619286+00:00",
  "error_file_id": null,
  "expired_at": null,
  "expires_at": "2024-07-20T17:33:29.1578141+00:00",
  "failed_at": null,
  "finalizing_at": null,
  "id": "batch_e0a7ee28-82c4-46a2-a3a0-c13b3c4e390b",
  "in_progress_at": null,
  "input_file_id": "file-c55ec4e859d54738a313d767718a2ac5",
  "errors": null,
  "metadata": null,
  "object": "batch",
  "output_file_id": null,
  "request_counts": {
    "total": null,
    "completed": null,
    "failed": null
  },
  "status": "Validating"
}

The following status values are possible:

Status Description
validating The input file is being validated before the batch processing can begin.
failed The input file has failed the validation process.
in_progress The input file was successfully validated and the batch is currently running.
finalizing The batch has completed and the results are being prepared.
completed The batch has been completed and the results are ready.
expired The batch was not able to be completed within the 24-hour time window.
cancelling The batch is being cancelled (This can take up to 10 minutes to go into effect.)
cancelled the batch was cancelled.

Retrieve batch job output file

curl https://YOUR_RESOURCE_NAME.openai.azure.com/openai/files/{output_file_id}/content?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY" > batch_output.jsonl

Additional batch commands

Cancel batch

Cancels an in-progress batch. The batch will be in status cancelling for up to 10 minutes, before changing to cancelled, where it will have partial results (if any) available in the output file.

curl https://YOUR_RESOURCE_NAME.openai.azure.com/openai/batches/{batch_id}/cancel?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY"

List batch

List existing batch jobs for a given Azure OpenAI resource.

curl https://YOUR_RESOURCE_NAME.openai.azure.com/openai/batches?api-version=2024-10-21 \
  -H "api-key: $AZURE_OPENAI_API_KEY"

The list API call is paginated. The response contains a boolean has_more to indicate when there are more results to iterate through.

List batch (Preview)

Use the REST API to list all batch jobs with additional sorting/filtering options.

curl "YOUR_RESOURCE_NAME.openai.azure.com/batches?api-version=2024-10-01-preview&$filter=created_at%20gt%201728773533%20and%20created_at%20lt%201729032733%20and%20status%20eq%20'Completed'&$orderby=created_at%20asc" \
  -H "api-key: $AZURE_OPENAI_API_KEY"

To avoid the error URL rejected: Malformed input to a URL function spaces are replaced with %20.

Global batch limits

Limit Name Limit Value
Max files per resource 500
Max input file size 200 MB
Max requests per file 100,000

Global batch quota

The table shows the batch quota limit. Quota values for global batch are represented in terms of enqueued tokens. When you submit a file for batch processing the number of tokens present in the file are counted. Until the batch job reaches a terminal state, those tokens will count against your total enqueued token limit.

Model Enterprise agreement Default Monthly credit card based subscriptions MSDN subscriptions Azure for Students, Free Trials
gpt-4o 5 B 200 M 50 M 90 K N/A
gpt-4o-mini 15 B 1 B 50 M 90 K N/A
gpt-4-turbo 300 M 80 M 40 M 90 K N/A
gpt-4 150 M 30 M 5 M 100 K N/A
gpt-35-turbo 10 B 1 B 100 M 2 M 50 K

B = billion | M = million | K = thousand

Batch object

Property Type Definition
id string
object string batch
endpoint string The API endpoint used by the batch
errors object
input_file_id string The ID of the input file for the batch
completion_window string The time frame within which the batch should be processed
status string The current status of the batch. Possible values: validating, failed, in_progress, finalizing, completed, expired, cancelling, cancelled.
output_file_id string The ID of the file containing the outputs of successfully executed requests.
error_file_id string The ID of the file containing the outputs of requests with errors.
created_at integer A timestamp when this batch was created (in unix epochs).
in_progress_at integer A timestamp when this batch started progressing (in unix epochs).
expires_at integer A timestamp when this batch will expire (in unix epochs).
finalizing_at integer A timestamp when this batch started finalizing (in unix epochs).
completed_at integer A timestamp when this batch started finalizing (in unix epochs).
failed_at integer A timestamp when this batch failed (in unix epochs)
expired_at integer A timestamp when this batch expired (in unix epochs).
cancelling_at integer A timestamp when this batch started cancelling (in unix epochs).
cancelled_at integer A timestamp when this batch was cancelled (in unix epochs).
request_counts object Object structure:

total integer
The total number of requests in the batch.
completed integer
The number of requests in the batch that have been completed successfully.
failed integer
The number of requests in the batch that have failed.
metadata map A set of key-value pairs that can be attached to the batch. This property can be useful for storing additional information about the batch in a structured format.

Frequently asked questions (FAQ)

Can images be used with the batch API?

This capability is limited to certain multi-modal models. Currently only GPT-4o support images as part of batch requests. Images can be provided as input either via image url or a base64 encoded representation of the image. Images for batch are currently not supported with GPT-4 Turbo.

Can I use the batch API with fine-tuned models?

This is currently not supported.

Can I use the batch API for embeddings models?

This is currently not supported.

Does content filtering work with Global Batch deployment?

Yes. Similar to other deployment types, you can create content filters and associate them with the Global Batch deployment type.

Can I request additional quota?

Yes, from the quota page in the Studio UI. Default quota allocation can be found in the quota and limits article.

What happens if the API doesn't complete my request within the 24 hour time frame?

We aim to process these requests within 24 hours; we don't expire the jobs that take longer. You can cancel the job anytime. When you cancel the job, any remaining work is cancelled and any already completed work is returned. You'll be charged for any completed work.

How many requests can I queue using batch?

There's no fixed limit on the number of requests you can batch, however, it will depend on your enqueued token quota. Your enqueued token quota includes the maximum number of input tokens you can enqueue at one time.

Once your batch request is completed, your batch rate limit is reset, as your input tokens are cleared. The limit depends on the number of global requests in the queue. If the Batch API queue processes your batches quickly, your batch rate limit is reset more quickly.

Troubleshooting

A job is successful when status is Completed. Successful jobs will still generate an error_file_id, but it will be associated with an empty file with zero bytes.

When a job failure occurs, you'll find details about the failure in the errors property:

"value": [
        {
          "id": "batch_80f5ad38-e05b-49bf-b2d6-a799db8466da",
          "completion_window": "24h",
          "created_at": 1725419394,
          "endpoint": "/chat/completions",
          "input_file_id": "file-c2d9a7881c8a466285e6f76f6321a681",
          "object": "batch",
          "status": "failed",
          "cancelled_at": null,
          "cancelling_at": null,
          "completed_at": 1725419955,
          "error_file_id": "file-3b0f9beb-11ce-4796-bc31-d54e675f28fb",
          "errors": {
                "object": “list”,
                "data": [
                {
               “code”: “empty_file”,
               “message”: “The input file is empty. Please ensure that the batch contains at least one   request.”
                    }
                ]
          },
          "expired_at": null,
          "expires_at": 1725505794,
          "failed_at": null,
          "finalizing_at": 1725419710,
          "in_progress_at": 1725419572,
          "metadata": null,
          "output_file_id": "file-ef12af98-dbbc-4d27-8309-2df57feed572",

            "request_counts": {
                "total": 10,
                "completed": null,
                "failed": null
            },
        }

Error codes

Error code Definition
invalid_json_line A line (or multiple) in your input file wasn't able to be parsed as valid json.

Please ensure no typos, proper opening and closing brackets, and quotes as per JSON standard, and resubmit the request.
too_many_tasks The number of requests in the input file exceeds the maximum allowed value of 100,000.

Please ensure your total requests are under 100,000 and resubmit the job.
url_mismatch Either a row in your input file has a URL that doesn’t match the rest of the rows, or the URL specified in the input file doesn’t match the expected endpoint URL.

Please ensure all request URLs are the same, and that they match the endpoint URL associated with your Azure OpenAI deployment.
model_not_found The Azure OpenAI model deployment name that was specified in the model property of the input file wasn't found.

Please ensure this name points to a valid Azure OpenAI model deployment.
duplicate_custom_id The custom ID for this request is a duplicate of the custom ID in another request.
empty_batch Please check your input file to ensure that the custom ID parameter is unique for each request in the batch.
model_mismatch The Azure OpenAI model deployment name that was specified in the model property of this request in the input file doesn't match the rest of the file.

Please ensure that all requests in the batch point to the same AOAI model deployment in the model property of the request.
invalid_request The schema of the input line is invalid or the deployment SKU is invalid.

Please ensure the properties of the request in your input file match the expected input properties, and that the Azure OpenAI deployment SKU is globalbatch for batch API requests.

Known issues

  • Resources deployed with Azure CLI won't work out-of-box with Azure OpenAI global batch. This is due to an issue where resources deployed using this method have endpoint subdomains that don't follow the https://your-resource-name.openai.azure.com pattern. A workaround for this issue is to deploy a new Azure OpenAI resource using one of the other common deployment methods which will properly handle the subdomain setup as part of the deployment process.

See also