AI summarization not working after upgrading to 3.3.0 beta1

@falco, continuing this from the last topic since it auto closed. Unfortunately after upgrading to 3.3.0 beta1, the AI Summarization has stopped working completely, as in the button to summarize has completely disappeared. I can only see the buttons on topics that were previously summarized with AI but no longer on any new topics (I’ve checked the topic summarization threshold).

When I look at the logs I see this error, this is new and wasn’t there earlier:

Message

DiscourseAi::Completions::Endpoints::Gemini: status: 429 - body: {
  "error": {
    "code": 429,
    "message": "Quota exceeded for quota metric 'Generate Content API requests per minute' and limit 'GenerateContent request limit per minute for a region' of service 'generativelanguage.googleapis.com' for consumer 'project_number:184805424415'.",
    "status": "RESOURCE_EXHAUSTED",
    "details": [
      {
        "@type": "type.googleapis.com/google.rpc.ErrorInfo",
        "reason": "RATE_LIMIT_EXCEEDED",
        "domain": "googleapis.com",
        "metadata": {
          "quota_limit_value": "0",
          "quota_location": "us-east2",
          "quota_metric": "generativelanguage.googleapis.com/generate_content_requests",
          "service": "generativelanguage.googleapis.com",
          "quota_limit": "GenerateContentRequestsPerMinutePerProjectPerRegion",
          "consumer": "projects/184805424415"
        }
      },
      {
        "@type": "type.googleapis.com/google.rpc.Help",
        "links": [
          {
            "description": "Request a higher quota limit.",
            "url": "https://cloud.google.com/docs/quota#requesting_higher_quota"
          }
        ]
      }
    ]
  }
}


Backtrace

/var/www/discourse/plugins/discourse-ai/lib/completions/endpoints/base.rb:93:in `block (2 levels) in perform_completion!'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/net-http-0.4.1/lib/net/http.rb:2353:in `block in transport_request'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/net-http-0.4.1/lib/net/http/response.rb:320:in `reading_body'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/net-http-0.4.1/lib/net/http.rb:2352:in `transport_request'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/net-http-0.4.1/lib/net/http.rb:2306:in `request'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/rack-mini-profiler-3.3.1/lib/patches/net_patches.rb:19:in `block in request_with_mini_profiler'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/rack-mini-profiler-3.3.1/lib/mini_profiler/profiling_methods.rb:50:in `step'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/rack-mini-profiler-3.3.1/lib/patches/net_patches.rb:18:in `request_with_mini_profiler'
/var/www/discourse/plugins/discourse-ai/lib/completions/endpoints/base.rb:91:in `block in perform_completion!'
/var/www/discourse/vendor/bundle/ruby/3.2.0/gems/net-http-0.4.1/lib/net/http.rb:1570:in `start'

Env

HTTP HOSTS: xxx.com

This is also set to 50: "Net::HTTPBadResponse" errors on Gemini Embeddings - #2 by Falco

Is there anything I can provide to see what’s going on?

Looks like it’s a very well described error, they even gave you a place to request a quota increase.

1 Like

I guess my question is why am I hitting a rate limit error when I’ve set the limit to 50 requests per minute (Gemini has a limit of 60 per minute) using the instructions you had provided earlier?

Those will limit the automated background calls, but you need to sum the user initiated calls on top of that. For more information check the logs on GCP.

1 Like

Thanks. Right now I’m trying to limit access to AI to just the staff. I’ve double checked all the AI settings and limited the trust levels to staff where available.

I’m guessing there are some modules which dont have trust limits (Search? Summarization?). Can crawlers or users who are not logged in invoke these AI features? It’s a small site so I know it isn’t logged in users who are using the API’s.

Also, I’m not able to understand why the summarize with AI button has disappeared and how to get it back. Any insight here would be appreciated.

FWIW, Which groups can use the AI to summarise topics is controlled by custom summarization allowed groups. You could check that setting and see if you have the groups you’re expecting in there?

I think for Semantic Search the AI power is within creating the embeddings, so once that is done people can use the AI Search without further calls on the AI provider itself. Now I’m not sure that’s accurate.

1 Like

To round this out, there was something wrong with Gemini. The quotas page is blank and does not show any metrics against the API key.

In-short solution was to delete the API key and regenerate a new one. Now I can see the metrics usage for the new API key/project and it’s working fine. Thanks Falco.

2 Likes

Some additional feedback here. Once the key stopped working, I replaced it with a new key but the AI features did not work, the BOT helper (Forum and Settings Explorer - no response from either and errors in the logs) didn’t work, the AI summarization button disappeared etc.

I tried to disable/enable the AI module from the admin settings, stop and start the container using ./launcher stop app but that didn’t help either. I needed to restart the entire machine the AI functions to start working again.

EDIT: It appears that setting the custom summarization allowed groups to Everyone disables the AI Summarization button for everyone (including staff) @JammyDodger

Yeah, that everyone pseudo-group is a tricksy one. I think there are plans to remove it from those type of settings soon to prevent it misleading more people. TL0 is the group that would work for this.

TL0 doesn’t work for users who aren’t logged-in. So if we want to let anonymous users use the Summarize AI option, it won’t work with TL0.

I think these features are currently for logged-in members only.

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.