# Documentation - Class name: LLMHtmlComposer - Category: SALT/Language Toolkit/Tools/HTML - Output node: False LLMHtmlComposer节点旨在根据给定的文本输入、分类器列表和可选的额外指令生成HTML内容。它利用语言模型来组合完整的HTML页面文档或HTML片段,具体取决于指定的模式,确保输出结构良好且有效。 # Input types ## Required - llm_model - 表示用于生成HTML内容的语言模型。它对于解释输入文本和分类器以产生所需的HTML输出至关重要。 - Comfy dtype: LLM_MODEL - Python dtype: dict - text_input - 用于生成HTML内容的主要文本输入。这段文本作为composition过程的基础。 - Comfy dtype: STRING - Python dtype: str - classifier_list - 以逗号分隔的字符串形式提供的分类器列表。这些分类器指导语言模型生成符合特定标准或类别的HTML内容。 - Comfy dtype: STRING - Python dtype: str ## Optional - extra_directions - 可选的附加指令,用于进一步指导HTML生成过程,允许生成更加定制化的输出。 - Comfy dtype: STRING - Python dtype: str - composer_mode - 指定HTML composition的模式,可以是完整的HTML页面文档或HTML片段,影响输出中HTML元素的结构和包含。 - Comfy dtype: COMBO[STRING] - Python dtype: str # Output types - html_output - 生成的HTML内容,可以是完整的HTML页面或HTML片段,基于输入文本、分类器和可选指令。 - Comfy dtype: STRING - Python dtype: str ## Usage tips - Infra type: `GPU` - Common nodes: unknown ## Source code ```python class LLMHtmlComposer: @classmethod def INPUT_TYPES(cls): return { "required": { "llm_model": ("LLM_MODEL",), "text_input": ("STRING", {"multiline": True, "dynamicPrompts": False, "placeholder": "Data..."}), "classifier_list": ("STRING", {"multiline": False, "dynamicPrompts": False}), }, "optional": { "extra_directions": ("STRING", {"multiline": True, "dynamicPrompts": False, "placeholder": "Extra directions for the LLM to follow..."}), "composer_mode": (["full_markup", "blocked_markup"],) } } RETURN_TYPES = ("STRING",) RETURN_NAMES = ("html_output",) FUNCTION = "compose_html" CATEGORY = f"{MENU_NAME}/{SUB_MENU_NAME}/Tools/HTML" def compose_html(self, llm_model, text_input, classifier_list, extra_directions="", composer_mode="full_markup"): classifier_list = [item.strip() for item in classifier_list.split(",") if item.strip()] markup_style = "full HTML page document" if composer_mode == "full_markup" else "HTML snippet (without html, head, body or any extraneous containers)" prompt = ( f"{text_input}\n\n###\n\n" f"Given the above text, create a valid {markup_style} utilizing *all* of the data, intact; " f"using the following classifiers: {classifier_list}.\n\n" f"{extra_directions}\n\n" "Please ensure the HTML output is well-structured, valid, and does not omit any data." ) response = llm_model['llm'].complete(prompt) return (response.text,) ```