feat: add a converter for prompt-template

[tikikun]

Refer to #754

Problem
Currently inference engine is doing their own way formatting user chat using their own way of templating the chat, this make the normal chat template not compatible with popular chat tempalte

Success Criteria
Need to write a few simple functions to convert from popular template -> template for supported engine to parse prompt_template option

[tikikun]

.

[tikikun]

cc @vuonghoainam also please help qa the function @hahuyhoang411

function splitString(promptTemplate) {
    // Split the string using the markers
    const systemMarker = "{system_message}";
    const promptMarker = "{prompt}";

    if (promptTemplate.includes(systemMarker) && promptTemplate.includes(promptMarker)) {
        // Find the indices of the markers
        const systemIndex = promptTemplate.indexOf(systemMarker);
        const promptIndex = promptTemplate.indexOf(promptMarker);

        // Extract the parts of the string
        const system_prompt = promptTemplate.substring(0, systemIndex);
        const user_prompt = promptTemplate.substring(systemIndex + systemMarker.length, promptIndex);
        const ai_prompt = promptTemplate.substring(promptIndex + promptMarker.length);

        // Return the split parts
        return { system_prompt, user_prompt, ai_prompt };
    } else if (promptTemplate.includes(promptMarker)) {
        // Extract the parts of the string for the case where only promptMarker is present
        const promptIndex = promptTemplate.indexOf(promptMarker);
        const user_prompt = promptTemplate.substring(0, promptIndex);
        const ai_prompt = promptTemplate.substring(promptIndex + promptMarker.length);

        // Return the split parts
        return { user_prompt, ai_prompt };
    }

    // Return an error if none of the conditions are met
    return { error: "Cannot split" };
}

[tikikun]

ideally it should split the bloke format to normal format but still containing some edge case @hahuyhoang411 can use in your own model list usage to see where it broke

[hahuyhoang411]

Tested with all of our models in the Hub. Thanks for the fix.