Librechat deployed on kubernetes its outputs is not full.

[Chethanibms]

Librechat deployed on kubernetes its outputs in not full, every time we need press the continue button.
attaching screenshot for your reference.

[danny-avila]

This is not the norm.

Usually this happens when “max_tokens” is a lower value, but no idea why this would be the case for you. It could be the fact that you’re using “gpt-4-vision-preview.” Or some predefined value to the Azure deployment.

try using another model or increasing the value for “Max Output Tokens” in the conversation settings here:

[danny-avila]

This seems to be a recent change with Azure OpenAI: #2884

Troubleshooting this.

[Chethanibms]

This is my configmap looks like, Please suggest if this is correct

apiVersion: v1
data:
  librechat.yaml: |
    version: 1.1.0
    cache: true
    endpoints:
      azureOpenAI:
        titleModel: "gpt-4-vision-preview"
        titleConvo: true
        plugins: true
        groups:
        - group: "azure-openai"
          apiKey: *****
          instanceName: "proview-vision"
          deploymentName: proview-vision
          version: "2024-02-15-preview"
          models:
            gpt-4-vision-preview:
              maxContextTokens: 128000
              max_tokens: 4096
              temperature: 0.2
              deploymentName: proview-vision

[danny-avila]

        gpt-4-vision-preview:
         maxContextTokens: 128000
         max_tokens: 4096

These fields won't actually do anything, just FYI.

I've issued a workaround for this specific issue with gpt-4-vision-preview, so you would just need to update to the latest build

[Chethanibms]

Means the Librechat version are the gpt-4-vision-preview modle deployment.

[Chethanibms]

Hi @danny-avila Images upload are not supporting in gpt-4-turbo-2024-04-09 model, Do we need to change any configurations.

[Chethanibms]

[danny-avila]

It's helpful to check your logs, which are usually found in the ./logs directory of your project root, or however you can view them in your kubernetes setup.

For example, I find the same error because maybe the "gpt-4-turbo" version I have is not the vision capable one (either dependent on the API version or the exact deployment I chose)

2024-05-30T13:16:41.188Z debug: Making request to https://{my_instance}.openai.azure.com/openai/deployments/gpt-4-1106-preview/chat/completions?api-version=2024-03-01-preview
2024-05-30T13:16:42.012Z warn: [OpenAIClient.chatCompletion][stream] API error
2024-05-30T13:16:42.014Z error: [handleAbortError] AI response error; aborting request: 400 Invalid content type. image_url is only supported by certain models.
2024-05-30T13:16:42.024Z debug: [AskController] Request closed

As you can see, when I switch it to a valid deployment for images, it works fine:

2024-05-30T13:24:30.786Z debug: Making request to https://{my_instance}.openai.azure.com/openai/deployments/gpt-4-turbo-2024-04-09/chat/completions?api-version=2024-03-01-preview
2024-05-30T13:24:43.333Z debug: [OpenAIClient] chatCompletion response