Skip to main content
This guide explains how to perform batch predictions using TrueFoundry’s AI Gateway with OpenAI, Vertex AI, or AWS Bedrock providers.

​
Client Setup

All providers use the OpenAI SDK with provider-specific headers. Choose your provider to get started: 
from openai import OpenAI

client = OpenAI(
    api_key="your-truefoundry-api-key",
    base_url="https://{controlPlaneUrl}/api/llm",
    default_headers={
        "x-tfy-provider-name": "openai-main"  # truefoundry provider integration name
    }
)

​
Input File Format

Create a JSONL file with one JSON object per line. Each line represents a single request: 
{"custom_id": "request-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "gpt-3.5-turbo-0125", "messages": [{"role": "system", "content": "You are a helpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
{"custom_id": "request-2", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "gpt-3.5-turbo-0125", "messages": [{"role": "system", "content": "You are an unhelpful assistant."},{"role": "user", "content": "Hello world!"}],"max_tokens": 1000}}
{"custom_id": "request-3", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "gpt-4-vision-preview", "messages": [{"role": "user", "content": [{"type": "text", "text": "What's in this image?"}, {"type": "image_url", "image_url": {"url": "https://example.com/image.jpg"}}]}], "max_tokens": 1000}}
Requirements: Valid JSON per line, meaningful custom_id values, minimum 100 records for AWS Bedrock.
Before using AWS Bedrock batch processing, ensure you have:
  • S3 Bucket: For storing input and output files
  • IAM Execution Role: With permissions for S3 access and Bedrock model invocation
  • User Permissions: Including iam:PassRole to pass the execution role to Bedrock
Before using Vertex AI batch processing, ensure you have:
  • Cloud Storage Bucket: The bucket must be in the same region as the model, and the service account must have read/write access to the bucket
  • Batch Prediction Permissions: The service account must have permission to create batch prediction jobs

​
Workflow Steps

The batch process follows these steps for all providers:
  1. Upload: Upload JSONL file → Get file ID
  2. Create: Create batch job → Get batch ID
  3. Monitor: Check status until complete
  4. Fetch: Download results

​
Step-by-Step Examples

from openai import OpenAI

client = OpenAI(
    api_key="your-truefoundry-api-key",
    base_url="https://{controlPlaneUrl}/api/llm",
    default_headers={
        "x-tfy-provider-name": "openai-main"  # truefoundry provider integration name
    }
)

# Upload the input file
file = client.files.create(
    file=open("request.jsonl", "rb"),
    purpose="batch"
)

print(file.id)  # Example: file-PnFGrFLN5LjjcWr4eFsStK

from openai import OpenAI

client = OpenAI(
    api_key="your-truefoundry-api-key",
    base_url="https://{controlPlaneUrl}/api/llm",
    default_headers={
        "x-tfy-provider-name": "openai-main"  # truefoundry provider integration name
    }
)

batch_job = client.batches.create(
    input_file_id=file.id,
    endpoint="/v1/chat/completions",
    completion_window="24h"
)

print(batch_job.id)  # Example: batch_67f7bfc50b288190893f242d9fa47c52

from openai import OpenAI

client = OpenAI(
    api_key="your-truefoundry-api-key",
    base_url="https://{controlPlaneUrl}/api/llm",
    default_headers={
        "x-tfy-provider-name": "openai-main"  # truefoundry provider integration name
    }
)

batch_status = client.batches.retrieve(batch_job.id)
print(batch_status.status)  # Example: completed, validating, in_progress, etc.

from openai import OpenAI

client = OpenAI(
    api_key="your-truefoundry-api-key",
    base_url="https://{controlPlaneUrl}/api/llm",
    default_headers={
        "x-tfy-provider-name": "openai-main"  # truefoundry provider integration name
    }
)

if batch_status.status == "completed":
    output_content = client.files.content(batch_status.output_file_id)
    print(output_content.content.decode('utf-8'))

​
Batch Status Reference

  • validating: Initial validation of the batch
  • in_progress: Processing the requests
  • completed: All requests processed successfully
  • failed: Batch processing failed

​
Best Practices

  1. File Format: Use meaningful custom_id values and valid JSONL format
  2. Error Handling: Implement proper error handling and status monitoring
  3. Security: Store API keys securely, use minimal IAM permissions
  4. AWS Bedrock Specific:
    • Minimum 100 records required in JSONL file
    • Verify IAM roles and S3 bucket permissions

​
Vertex AI Permissions

The following permissions are required for Vertex AI batch prediction:
Cloud Storage bucket
  • The Cloud Storage bucket used for batch input and output must be in the same region as the Vertex AI model you are using.
  • The service account that runs the batch job must have read and write access to this bucket (e.g. roles/storage.objectAdmin or equivalent object-level read/write permissions on the bucket).
Create batch prediction jobs
  • The service account must have permission to create and manage batch prediction jobs in Vertex AI (e.g. roles/aiplatform.user or the aiplatform.batchPredictionJobs.create permission).

​
AWS Bedrock Permissions

These are the minimum permissions required to use the Bedrock Batch APIs. For complete official guidance, see AWS Bedrock Batch Inference Permissions.
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "bedrock:ListFoundationModels",
        "bedrock:GetFoundationModel",
        "bedrock:ListInferenceProfiles",
        "bedrock:GetInferenceProfile",
        "bedrock:ListCustomModels",
        "bedrock:GetCustomModel",
        "bedrock:TagResource",
        "bedrock:UntagResource",
        "bedrock:ListTagsForResource",
        "bedrock:CreateModelInvocationJob",
        "bedrock:GetModelInvocationJob",
        "bedrock:ListModelInvocationJobs",
        "bedrock:StopModelInvocationJob"
      ],
      "Resource": [
        "arn:aws:bedrock:<region>:<account_id>:model-customization-job/*",
        "arn:aws:bedrock:<region>:<account_id>:custom-model/*",
        "arn:aws:bedrock:<region>::foundation-model/*"
      ]
    },
    {
      "Effect": "Allow",
      "Action": ["s3:ListBucket", "s3:PutObject", "s3:GetObject", "s3:GetObjectAttributes"],
      "Resource": ["arn:aws:s3:::<bucket>", "arn:aws:s3:::<bucket>/*"]
    },
    {
      "Action": ["iam:PassRole"],
      "Effect": "Allow",
      "Resource": "arn:aws:iam::<account_id>:role/<service_role_name>",
      "Condition": {
        "StringEquals": {
          "iam:PassedToService": ["bedrock.amazonaws.com"]
        }
      }
    }
  ]
}
The service role (role_arn) used for creating and executing the batch job requires:Trust Relationship:
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": {
        "Service": "bedrock.amazonaws.com"
      },
      "Action": "sts:AssumeRole",
      "Condition": {
        "StringEquals": {
          "aws:SourceAccount": "<account_id>"
        },
        "ArnEquals": {
          "aws:SourceArn": "arn:aws:bedrock:<region>:<account_id>:model-invocation-job/*"
        }
      }
    }
  ]
}
Permission Policy:
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": ["s3:GetObject", "s3:PutObject", "s3:ListBucket"],
      "Resource": ["arn:aws:s3:::<bucket>", "arn:aws:s3:::<bucket>/*"]
    }
  ]
}