Can I specify a model when use remote llama server?

[gitsang]

I have a llm provider use lm studio:

And I use this config:

  {
    "ggml-org/llama.vim",
    init = function()
      vim.g.llama_config = {
        endpoint = "http://10.60.20.7:1234/v1/completions",
      }
    end,
  }

I realized that there no way for me to specify a model.

{
    "error": {
        "message": "Multiple models are loaded. Please specify a model by providing a 'model' field in the request body.\n\nYour models:\n\ntext-embedding-qwen3-embedding-0.6b\ntext-embedding-qwen3-embedding-8b\nqwen3-32b-mlx\nqwen/qwen3-30b-a3b-2507",
        "type": "invalid_request_error",
        "param": "model",
        "code": "model_not_found"
    }
}

[liberlanco]

Hello. I've stuck with the same problem with 'llama-swap' frontend. I've researched 'llama.vscode' and there is a support for 'model' option. So I decided to add it here too - #81.

Also please note that you use /v1/completions endpoint. I'm not sure that it will work, docs recommends to use /infill endpoint.

[cristoper]

As a workaround for llama-swap (but not lmstudio unfortunately) until liberlanco's PR gets merged, you can use the /upstream endpoint to configure llama.vim to talk to a specific model. For example, I use let g:llama_config.endpoint = 'http://localhost:8100/upstream/qwen3-30b-a3b-coder-52k/infill'

[arb000r]

Can you use the model config merged via #81 to use an mlx model hosted via lmstudio? The /v1/completions endpoint doesn't seem to do much