vlm-1 can extract structured data from invoices, along with their visual grounding in PDF or image format. Here’s a step-by-step guide on how to parse an invoice: Here is a visualization of the parsed invoice along with the visual grounding that vlm-1 can extract from an invoice. Notice that only the specific items requested in the schema are retrieved and visualized, unlike OCR which returns all text in the document with no context:

Parsing an invoice with visual grounding enabled

For higher-quality results, we recommend enabling Visual Grounding to help the model understand the invoice and extract more accurate information. See High-Accuracy Parsing with Grounding for more details.

​
Parsing Invoices in 2 Steps

1

Submit an Invoice Parsing Job

from pathlib import Path
from vlmrun.client import VLMRun
from vlmrun.client.types import FileResponse

# Initialize the client
client = VLMRun(api_key="<your-api-key>")

# Submit the invoice for parsing
response: PredictionResponse = client.document.generate(
    file=Path("<path/to/invoice.pdf>"),
    domain="document.invoice",
    batch=True,
)
print(f"Job submitted:\n {response.model_dump()}")
You should see a response like this:
Job submitted:
{
  "id": "052cf2a8-2b84-45f5-a385-ccac2aae13bb",
  "created_at": "2024-08-15T02:22:09.157788",
  "response": null,
  "status": "pending"
}
2

Wait for the Job to Complete

You can now wait for the job to complete by calling the predictions.wait method:
# Wait for the job to complete
response: PredictionResponse = client.predictions.wait(
    id=response.id,
    timeout=120,
)
print(f"Job completed:\n {response.model_dump()}")
You should see a response like this: 
{
    "id": "052cf2a8-2b84-45f5-a385-ccac2aae13bb",
    "created_at": "2024-08-15T02:22:09.157788",
    "status": "completed",
    "response": {
    "currency": "USD",
    "currency_metadata": {
      "bboxes": [
        {
          "content": "$19,647.68",
          "bbox": {
            "xywh": [0.843, 0.611, 0.084, 0.014]
          },
          "page": 0
        }
      ]
    },
    "customer": "Jane Smith",
    "customer_billing_address": {
      "city": "Mountain View",
      "city_metadata": {
        "bboxes": [
          {
            "content": "Mountain View, CA 94043",
            "bbox": {
              "xywh": [0.080, 0.194, 0.190, 0.014]
            },
            "page": 0
          }
        ]
      },
      "country": null,
      "country_metadata": null,
      "postal_code": "94043",
      "postal_code_metadata": {
        "bboxes": [
          {
            "content": "Mountain View, CA 94043",
            "bbox": {
              "xywh": [0.080, 0.194, 0.190, 0.014]
            },
            "page": 0
          }
        ]
      },
    }
    ...
    "items": [...],   // List of items in the invoice
    ...
    "total": 19647.68,
    "total_metadata": {
      "bboxes": [
        {
          "content": "$19,647.68",
          "bbox": {
            "xywh": [...]
          },
          "page": 0
        }
      ]
    }
  }
}

​
High-Accuracy Parsing with Grounding

For higher-quality results, you can enable Visual Grounding to help the model understand the invoice and extract more accurate information. You can do this by setting the config=GenerationConfig(grounding=True) parameter when submitting the job (as shown below). 
from vlmrun.client.types import GenerationConfig

# Enable grounding when submitting the job
response: PredictionResponse = client.document.generate(
    file=Path("<path/to/invoice.pdf>"),
    domain="document.invoice",
    batch=True,
    config=GenerationConfig(grounding=True),
)

​
Try our Document -> JSON API today

Head over to our Document -> JSON to start building your own document processing pipeline with VLM Run. Sign-up for access on our platform.