Extensions/comfyui_LLM_schools
ComfyUI Extension

comfyui_LLM_schools

ComfyUI node library for fine-tuning LLMs

By heshengtao·Created 2 years ago·Updated 2 years ago· 7
heshengtao/comfyui_LLM_schools
Nodes9
On cloudLocal install
Category大模型学校（llm_schools）/模型训练（Model Training）, 大模型学校（llm_schools）/数据预处理（data preprocessing）
Stars7
Updated2 years ago

Nodes (9)

Causal Language Model Trainer

Fine-tune a real LLM from a ComfyUI node — with a catch

大模型学校（llm_schools）/模型训练（Model Training）
download/load the HF dataset

Pull any Hugging Face dataset into your workflow without leaving the graph

大模型学校（llm_schools）/数据预处理（data preprocessing）
get HF dataset repo_id

Find the right HF dataset repo_id without opening a browser tab

大模型学校（llm_schools）/数据预处理（data preprocessing）
IA3 Arguments

IA3's settings, for when you want the cheapest adapter that still works

大模型学校（llm_schools）/模型训练（Model Training）
LLM Arguments

The 'what should this training run even be' knobs, in one node

大模型学校（llm_schools）/模型训练（Model Training）
LoRA or Adapter Arguments

LoRA hyperparameters for the trainer, minus the LoRA jargon spiral

大模型学校（llm_schools）/模型训练（Model Training）
P or Prompt Arguments

Prompt tuning args — where 'RANDOM vs TEXT' decides your whole run

大模型学校（llm_schools）/模型训练（Model Training）
Prefix Arguments

Prefix tuning's config dials, exposed as a node

大模型学校（llm_schools）/模型训练（Model Training）
split HF dataset

Turn a raw dataset download into train/val/test splits — automatically

大模型学校（llm_schools）/数据预处理（data preprocessing）
Readme

comfyui_LLM_schools

<div align="center"> <a href="./README.md"><img src="https://img.shields.io/badge/English-d9d9d9"></a> <a href="./README_ZH.md"><img src="https://img.shields.io/badge/简体中文-d9d9d9"></a> </div>

  1. Currently supports full fine-tuning of AutoModelForCausalLM classes in the transformer library, refer to workflow The_first_training