ethz-spylab / satml-llm-ctf

Code used to run the platform for the LLM CTF colocated with SaTML 2024
https://ctf.spylab.ai
MIT License
23 stars 5 forks source link

HTTPException: Status code: 500 (utility check) #37

Closed NivC closed 8 months ago

NivC commented 8 months ago

Dear Spylab,

Validating our final model we got:

request": {
      "model": "openai/gpt-3.5-turbo-1106",
      "api_keys": null,
      "small": false
    },
    "result": {
      "utility": 0.629,
      "threshold": 0.483,
      "passed": true,
      "additional_info": {
        "avg_share_of_failed_queries": 0.016,
        "sample_errors": [
          "HTTPException: Status code: 500, Detail: ('OpenAI API error: The server had an error while processing your request. Sorry about that! {\\n  \"error\": {\\n    \"message\": \"The server had an error while processing your request. Sorry about that!\",\\n    \"type\": \"server_error\",\\n    \"param\": null,\\n    \"code\": null\\n  }\\n}\\n 500 {\\'error\\': {\\'message\\': \\'The server had an

So it seems to pass as far as we can tell. Should be worried about the error?

Niv.

dpaleka commented 8 months ago

Yeah, our implementation had an issue with retrying failed queries affecting how long the endpoint takes, so we just report how many queries have failed upstream. If there's an error other than a OpenAI/TogetherAI API issue, then that's a cause for concern (for either the server or your defense, depending on the error). It seems you've successfully tested on 98.4% of the data and get utility much higher than the threshold, so no worry about this in particular.

NivC commented 7 months ago

Thanks!