feat(core): support configuring stop words in model config

[zwpaper]

fix #3188

[wsxiaoys]

Please also test the behavior locally (e.g with screen recording). Otherwise LGTM

[wsxiaoys]

Suggested change

	pub stop_words: Vec<String>,
	pub additional_stop_words: Vec<String>,

[wsxiaoys]

Suggested change

	model_stop_words: Vec<String>,
	stop_words_from_model_config: Vec<String>,

[zwpaper]

when using the following config, I can reproduce the error:

[model.completion.http]
kind = "ollama/completion"
model_name = "qwen2.5-coder:7b-base"
api_endpoint = "http://localhost:11434"
prompt_template = "<|fim_prefix|>{prefix}<|fim_suffix|>{suffix}<|fim_middle|>"
stop_words = []

but some of the stop words seems to be not working, even with the following config:

[model.completion.http]
kind = "ollama/completion"
model_name = "qwen2.5-coder:7b-base"
api_endpoint = "http://localhost:11434"
prompt_template = "<|fim_prefix|>{prefix}<|fim_suffix|>{suffix}<|fim_middle|>"
stop_words = ["<|endoftext|>", "<|file_sep|>"]

---

still working on it and trying to figure out why

[wsxiaoys]

Suggested change

	let stop_condition_factory = match config {
	Some(ModelConfig::Local(config)) => StopConditionFactory::with_stop_words(
	config.additional_stop_words.unwrap_or_default(),
	),
	Some(ModelConfig::Http(config)) => StopConditionFactory::with_stop_words(
	config.additional_stop_words.unwrap_or_default(),
	),
	_ => StopConditionFactory::default(),
	};
	let additional_stop_words = match config {
	Some(ModelConfig::Local(config)) => config.additional_stop_words.unwrap_or_default(),
	...
	};

[zwpaper]

the stop_condition_factory here means all the stop words, not only the additional stop words, are we really want to rename the stop_condition_factory to additional_stop_words?

[wsxiaoys]

I'm asking a restructure, rather than a renaming.

        let additional_stop_words = match config {
            Some(ModelConfig::Local(config)) =>  config.additional_stop_words.unwrap_or_default(),
            ...
        };

       let stop_condition_factory = StopConditionFactory::with_stop_words(additional_stop_words).

So you don't need have to write StopConditionFactory::with_stop_words multiple times.

[wsxiaoys]

why change this?

[zwpaper]

Because the config passes in will be used later, so this must change to a borrowed args, or it will take over the life time of the config passed in

[wsxiaoys]

I recommend just clone it in this particular case.

[wsxiaoys]

Please fix unit test