Query with no results returns ApiError instead of an empty fallback when metadata filters are applied

[alexander-zuev]

Hey there 👋,

I've submitted this over support, but sharing more details here.

I am really looking forward to integrating Vectara into my app but unfortunately SDK (API under the hood) seems to return an error response in case metadata filters yield no results.

See this (dummy example):

• When using a metadata filter
  "metadata_filter": "doc.user_id = '58f128ae-c2fa-4a02-a0b5-4f83258e3e'"
• I receive this response

<Response [412]>
{"messages":["FAILED_PRECONDITION: Factual consistency computation skipped due to summarizer error."],"request_id":"70248967cd6771e2e4cc0ee6fa45e094"}

If I turn off metadata filter and send the exact same query + search params:
The response is

<Response [200]>
{"summary":"The returned results did not contain sufficient information..}

Here is the full code snippet I am using

`import json

import requests

url = "https://api.vectara.io/v2/corpora/kollektiv-mvp-dev/query"

payload = json.dumps(
    {
        "query": "Why did failed precondition error occur?",
        "search": {
            "custom_dimensions": {},
            # "metadata_filter": "doc.user_id = '58f128ae-c2fa-4a02-a0b5-4f83258e3e'",
            "lexical_interpolation": 0.025,
            "semantics": "default",
        },
        "generation": {
            "generation_preset_name": "vectara-summary-ext-v1.2.0",
            "max_used_search_results": 5,
            "max_response_characters": 300,
            "response_language": "eng",
            "citations": {
                "style": "none",
                "url_pattern": "https://vectara.com/documents/{doc.id}",
                "text_pattern": "{doc.title}",
            },
            "enable_factual_consistency_score": True,
        },
        "intelligent_query_rewriting": False,
    }
)
headers = {
    "Content-Type": "application/json",
    "Accept": "application/json",
    "x-api-key": "api-key",
}

response = requests.request("POST", url, headers=headers, data=payload)

print(response)
print(response.text)`

The same problem occurs with both SDK and API.

Also do note that
"enable_factual_consistency_score": False, - seems to affect nothing, the same error is produced when I set this to False or comment out the setting

If you'd like to me share more info / help debug - let me know!

Thanks!

[ofermend]

Hi @alexander-zuev - thanks for reporting this. We are working to fix this and apologies for the issue - should be fixed soon.
Separately, I see you are using "vectara-summary-ext-v1.2.0" in your query, please consider some of the alternative summarizers that might provide better results like "vectara-summary-table-query-ext-dec-2024-gpt-3-5" (assuming you want to keep GPT-3.5). Full list: https://docs.vectara.com/docs/learn/grounded-generation/select-a-summarizer

[alexander-zuev]

Awesome looking forward to it!

Thanks for the tip regarding the summarizer - will ensure to switch to more modern ones