[BUG]: 404 Error for nemoretriever-table-structure-v1 API Endpoint

[ansh-info]

Version

nemoretriever-table-structure-v1:latest(2025-03-17)

Describe the bug.

What I Tried

I attempted to call the following NVIDIA API endpoint:

POST https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-table-structure-v1

With this payload format:

{
  "input": [
    {
      "type": "image_url",
      "url": "data:image/png;base64,<base64_encoded_image>"
    }
  ]
}

And appropriate authorization headers:

Authorization: Bearer <MY_API_KEY>
Accept: application/json

What Happened

I consistently receive a 404 error:

{
  "status": 404,
  "title": "Not Found",
  "detail": "Function '2befd4f2-fe18-4a45-bc49-556bd74646b9': Not found for account 'xvrF-qIqK9kgyWlaCjOgCuw035tJBvpnCiPXj-TavH0'"
}

What I Expected

Since this endpoint is documented ([source: NVIDIA Foundation Model API documentation]), I expected it to process the request and return a table structure extraction result.

Notes

• My API key is valid (other models like paddleocr, nemoretriever-graphic-elements-v1 work).
• This suggests the model is not available for my account or project.
• I couldn't find a public endpoint to list available models (/v1/functions returns 404 as well).

---

Question

• Is nemoretriever-table-structure-v1 restricted to specific accounts or under private preview?
• If so, how can I request access to use this function?
• Is there an alternative model for table structure detection that is publicly available via the NVIDIA API?

Thanks for your help!

Minimum reproducible example

def detect_table_structure(image_pil, size=(896, 896), quality=85):
    import base64
    from io import BytesIO
    import requests
    import os

    # Resize and convert to JPEG
    img = image_pil.convert("RGB").resize(size)
    buffered = BytesIO()
    img.save(buffered, format="JPEG", quality=quality, optimize=True)
    image_b64 = base64.b64encode(buffered.getvalue()).decode()

    if len(image_b64) > 180_000:
        raise ValueError(f"Image too large: {len(image_b64)} chars")

    headers = {
        "Authorization": f"Bearer {os.getenv('NVIDIA_API_KEY')}",
        "Accept": "application/json"
    }

    payload = {
        "input": [
            {
                "type": "image_url",
                "url": f"data:image/jpeg;base64,{image_b64}"
            }
        ]
    }

    response = requests.post(
        "https://ai.api.nvidia.com/v1/cv/nvidia/nemoretriever-table-structure-v1",
        headers=headers,
        json=payload
    )

    response.raise_for_status()
    return response.json()

table_regions = [r for r in cropped_regions if r["type"] == "table"]
print(f"Found {len(table_regions)} table regions.")

# Try the first one
table_test = table_regions[0]
structure_result = detect_table_structure(table_test["image"], size=(640, 640), quality=70)
print(structure_result)

Relevant log output

{
  "status": 404,
  "title": "Not Found",
  "detail": "Function '2befd4f2-fe18-4a45-bc49-556bd74646b9': Not found for account 'xvrF-qIqK9kgyWlaCjOgCuw035tJBvpnCiPXj-TavH0'"
}

Full env printout

Other/Misc.

No response

Code of Conduct

• I agree to follow THIS PROJECT's Code of Conduct
• I have searched the open bugs and have found no duplicates for this bug report

[sumitkbh]

Do you still see this issue?

[ansh-info]

Do you still see this issue?

Its seems to work fine, Probably fixed - Thank You!!