Add support for base_url option to use local models

For example, you can start llama-cpp-python like this (it emulates
the openai api):
```sh
CMAKE_ARGS="-DLLAMA_CUBLAS=on" pip install 'llama-cpp-python[server]'
wget https://huggingface.co/TheBloke/CodeLlama-13B-Instruct-GGUF/resolve/main/codellama-13b-instruct.Q5_K_M.gguf
python3 -m llama_cpp.server --n_gpu_layers 100 --model codellama-13b-instruct.Q5_K_M.gguf
```

Then set the API url in your `.vimrc`:

```vim
let g:vim_ai_chat = {
    \ "engine": "chat",
    \ "options": {
        \ "base_url": "http://127.0.0.1:8000",
    \ },
\ }
```

And chat with the locally hosted AI using `:AIChat`.

The change in utils.py was needed because llama-cpp-python adds a new
line to the final response: `[DONE]^M`.

1 files changed, 4 insertions, 1 deletions

diff --git a/py/chat.py b/py/chat.py
index 93d31bf..7bca31d 100644
--- a/py/chat.py
+++ b/py/chat.py
@@ -1,3 +1,4 @@
+from urllib.parse import urljoin
 # import utils
 plugin_root = vim.eval("s:plugin_root")
 vim.command(f"py3file {plugin_root}/py/utils.py")
@@ -69,7 +70,9 @@ try:
             **openai_options
         }
         printDebug("[chat] request: {}", request)
-        response = openai_request('https://api.openai.com/v1/chat/completions', request, http_options)
+        base_url = options.get('base_url', 'https://api.openai.com')
+        url = urljoin(base_url, 'v1/chat/completions')
+        response = openai_request(url, request, http_options)
         def map_chunk(resp):
             printDebug("[chat] response: {}", resp)
             return resp['choices'][0]['delta'].get('content', '')