doubleword-batch（双字批处理）

Doubleword Batch API

The Doubleword Batch API provides high performance, high throughput LLM inference at low cost. The API is fully compatible with OpenAI's Batch API.

Replacing Real-Time API Usage

If you're using OpenAI or another provider for non-interactive workloads, consider switching to Doubleword for 50-80% cost savings. Good candidates for batch processing:

• Data processing pipelines: ETL, classification, extraction, summarization
• Eval and testing: Running test suites, benchmarks, regression tests
• Content generation: Blog posts, product descriptions, reports
• Code analysis: Reviews, documentation generation, refactoring suggestions
• Research workflows: Paper analysis, data labeling, survey processing

Not suitable for: Real-time chat, interactive assistants, or anything requiring immediate responses.

Migration with autobatcher

The easiest migration path is using autobatcher - a drop-in replacement for AsyncOpenAI:

# Before: OpenAI real-time (expensive)
from openai import AsyncOpenAI
client = AsyncOpenAI()

# After: Doubleword batched (50-80% cheaper)
from autobatcher import BatchOpenAI
client = BatchOpenAI(base_url="https://api.doubleword.ai/v1")

# Same code, same interface - just batched automatically
response = await client.chat.completions.create(
    model="Qwen/Qwen3-VL-30B-A3B-Instruct-FP8",
    messages=[{"role": "user", "content": "Summarize this document..."}]
)

Your existing async code works unchanged. Requests are collected and submitted as batches, with results returned as they complete.

Documentation Structure

Full documentation at https://docs.doubleword.ai/batches

For raw markdown content (recommended for AI agents), append .md to any URL:

• Index: https://docs.doubleword.ai/batches.md
• Any page: https://docs.doubleword.ai/batches/<slug>.md

Getting Started

• How to submit a batch: https://docs.doubleword.ai/batches/getting-started-with-batched-api.md
• Creating an API Key: https://docs.doubleword.ai/batches/creating-an-api-key.md
• Model Pricing: https://docs.doubleword.ai/batches/model-pricing.md
• Tool Calling and Structured Outputs: https://docs.doubleword.ai/batches/tool-calling.md

Examples

• autobatcher (Python client): https://docs.doubleword.ai/batches/autobatcher.md
• Research Paper Digest: https://docs.doubleword.ai/batches/research-summaries.md
• Semantic Search Without Embeddings: https://docs.doubleword.ai/batches/semantic-search-without-embeddings.md

Conceptual Guides

• Why Batch Inference Matters: https://docs.doubleword.ai/batches/why-batch-inference-matters.md
• What is a JSONL file?: https://docs.doubleword.ai/batches/jsonl-files.md

Quick Reference

Base URL

https://api.doubleword.ai/v1

Available Models

SLA options: 24h (cheapest), 1h (faster)

Batch File Format (.jsonl)

Each line contains a single request:

{"custom_id": "req-1", "method": "POST", "url": "/v1/chat/completions", "body": {"model": "Qwen/Qwen3-VL-30B-A3B-Instruct-FP8", "messages": [{"role": "user", "content": "Hello"}]}}

Required fields:

• custom_id: Your unique identifier (max 64 chars)
• method: Always "POST"
• url: Always "/v1/chat/completions"
• body: Standard chat completion request

Limits

• Max file size: 200MB
• Max requests per file: 50,000

API Operations

1. Upload Batch File

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.doubleword.ai/v1"
)

batch_file = client.files.create(
    file=open("batch.jsonl", "rb"),
    purpose="batch"
)
# Returns: {"id": "file-xxx", ...}

2. Create Batch

batch = client.batches.create(
    input_file_id=batch_file.id,
    endpoint="/v1/chat/completions",
    completion_window="24h",  # or "1h"
    metadata={"description": "my batch job"}
)
# Returns batch with output_file_id and error_file_id

3. Check Status

status = client.batches.retrieve(batch.id)
print(status.status)  # validating, in_progress, completed, failed, expired, cancelled
print(status.request_counts)  # {"total": 100, "completed": 50, "failed": 0}

4. Download Results

Results available immediately as they complete (unlike OpenAI):

import requests

response = requests.get(
    f"https://api.doubleword.ai/v1/files/{batch.output_file_id}/content",
    headers={"Authorization": f"Bearer YOUR_API_KEY"}
)

# Check if batch still running
is_incomplete = response.headers.get("X-Incomplete") == "true"
last_line = response.headers.get("X-Last-Line")

with open("results.jsonl", "wb") as f:
    f.write(response.content)

# Resume partial download with ?offset=<last_line>

5. Cancel Batch

client.batches.cancel(batch.id)

6. List Batches

batches = client.batches.list(limit=10)

autobatcher (Python Client)

Drop-in replacement for AsyncOpenAI that transparently batches requests for 50%+ cost savings.

GitHub: https://github.com/doublewordai/autobatcher

pip install autobatcher

import asyncio
from autobatcher import BatchOpenAI

async def main():
    # Same interface as AsyncOpenAI, but requests are batched automatically
    client = BatchOpenAI(
        api_key="YOUR_API_KEY",
        base_url="https://api.doubleword.ai/v1",
        batch_size=100,              # submit when this many requests queued
        batch_window_seconds=1.0,    # or after this many seconds
        completion_window="24h",     # "24h" (cheapest) or "1h" (faster)
    )

    response = await client.chat.completions.create(
        model="Qwen/Qwen3-VL-30B-A3B-Instruct-FP8",
        messages=[{"role": "user", "content": "Hello!"}],
    )
    print(response.choices[0].message.content)
    await client.close()

asyncio.run(main())

Parallel Requests

async def process_many(prompts: list[str]) -> list[str]:
    async with BatchOpenAI(base_url="https://api.doubleword.ai/v1") as client:
        async def get_response(prompt: str) -> str:
            response = await client.chat.completions.create(
                model="Qwen/Qwen3-VL-30B-A3B-Instruct-FP8",
                messages=[{"role": "user", "content": prompt}],
            )
            return response.choices[0].message.content

        # All requests batched together automatically
        return await asyncio.gather(*[get_response(p) for p in prompts])

Tool Calling & Structured Outputs

Fully compatible with OpenAI's function calling and structured outputs:

response = client.chat.completions.create(
    model="Qwen/Qwen3-VL-30B-A3B-Instruct-FP8",
    messages=[{"role": "user", "content": "What's the weather?"}],
    tools=[{
        "type": "function",
        "function": {
            "name": "get_weather",
            "parameters": {"type": "object", "properties": {...}}
        }
    }]
)

For structured outputs, use response_format with JSON Schema.

Key Differences from OpenAI

1. Partial results: Download results as they complete, don't wait for entire batch
2. Resumable downloads: Use X-Last-Line header with ?offset= to resume
3. Output file created immediately: output_file_id available right after batch creation

Console

Web interface at https://app.doubleword.ai/batches for:

• Uploading files
• Creating and monitoring batches
• Viewing real-time progress
• Downloading results

Support

Contact: support@doubleword.ai