OpenAI Configuration Seems to be Ignored

[beezu]

I've installed this plugin successfully using Lazy, and I'm trying to set it up using a locally hosted OpenAI-compatible API, hosted via LM Studio. This is the configuration I'm using:

return {
  {
    'huggingface/llm.nvim',
    opts = {
      {
        api_token = nil,
        backend = "openai",
        model = "qwen2.5-coder-32b-instruct",
        url = "http://localhost:1234/v1/completions",
        tokenizer = {
          path = "/Users/[USERNAME]/.tokenizer.json"
        },
        request_body = {
          temperature = 0.2,
          top_p = 0.95,
        }
      }
    }
  }
}

The plugin is not communicating with my self-hosted API at all, and is instead (successfully) using HuggingFace Inference API.