Default engine param not used for `chat`

[shannoncole]

I was surprised that the chat method doesn't use the configured default model.

Example:

ChatGPT.configure do |config|
  config.api_key = ENV["OPENAI_API_KEY"]
  config.api_version = "v1"
  config.default_engine = "gpt-4" # <--- default configured here
  config.request_timeout = 120
  config.default_parameters = {
    max_tokens: nil,
    temperature: 0.0,
    top_p: 0.3,
    n: 1,
  }
end

resp = client.chat(...) 
puts response.dig("model") # outputs: gpt-3.5-turbo

Relevant source:

    def chat(messages, params = {})
      url = "#{@endpoint}/chat/completions"

      data = @config.default_parameters.merge(
        model: params[:model] || 'gpt-3.5-turbo', # <--- hard-coded default
        messages: messages,
        temperature: params[:temperature],
        top_p: params[:top_p],
        n: params[:n],
        stream: params[:stream] || false
      ).compact

      request_api(url, data)
    end

I may have misunderstood what's meant to happen, but I expected the configured default_engine to be used instead.

Thanks for the library, I have found it very useful!

[nagstler]

@shannoncole

Thank you for pointing out this inconsistency!

In the current implementation, the chat and chat_stream methods have hardcoded defaults to gpt-3.5-turbo instead of using the configured default_engine

In chat method:

model: params[:model] || 'gpt-3.5-turbo',

While the completions method properly uses the configured value:

In completions method:

engine = params[:engine] || @config.default_engine

The fix will be straightforward - both chat methods should respect the configured default_engine value:

In chat method:

model: params[:model] || @config.default_engine,

I'll create a PR to fix it or feel free to raise a PR