I am writing a plugin, and I've understood that this will tell llm to always be in --no-stream mode when these models are called:

class Chat(llm.Model):
    can_stream: bool = False

Is this what's needed and all that is needed? The plugin will only run when stream=False to manage upstream user expectations.

def execute(self, prompt, stream, response, conversation):

When debugging that method, stream correctly ends up being false, even without --no-stream when not in chat mode.

But it seems that stream=True is always sent to execute() when in chat mode, and using --no-stream via command line will disable the chat mode itself. Here are some test runs with the plugin, vs. OpenAI gpt4 alias. The behavior seems similar.

> llm "hi"
Hello! How can I assist you today?
> llm chat
Error: Sorry, this plugin doesn't support streaming yet.     # stream was set to to True by the host
> llm --no-stream chat
Hello! How can I assist you today?     # chat mode not applied but there is a greeting
> llm chat -m gpt4
Chatting with gpt-4
Type 'exit' or 'quit' to exit
> llm --no-stream chat -m gpt4
Hello! How can I assist you today?     # chat mode not applied but there is a greeting
> llm chat -m gpt4 --no-stream
Usage: llm chat [OPTIONS]
Try 'llm chat --help' for help.
Error: No such option: --no-stream Did you mean --system?
> llm chat -o verbose     # the default model does have this option in non-chat mode
Usage: llm chat [OPTIONS]
Try 'llm chat --help' for help.
Error: No such option: -o

Applying the plugin's own announced streaming boolean would fix this for me. Allowing explicit --no-stream while in chat mode is nice but I don't personally have a strong use case to offer for that, other than perhaps consistency.

Do you plan to support model options in chat mode? They are often needed with local models. The new mode is a very welcome addition, so thanks a lot for working on it.