QBot: Clone your QQ friends

Overview

QBot is an pipeline for training large language models (LLMs) using QQ chat data. The project covers the full workflow: decrypting chat records, extracting and preprocessing data, LoRA fine-tuning, reward model training, RLHF, and Huggingface-based demos.

---

Quick Start

1. Install dependencies (TBD)

pip install -r requirements.txt

2. Decrypt QQ chat database

please quit your linux qq before runing script

python scripts/dump_qq_data.py

3. Extract chat data to JSON (interactive script)

python scripts/extract.py --profile_db data/raw/path-to-profile_info.decrypt.db --msg_db data/raw/path-to-nt_msg.decrypt.db

4. Preprocess data

python scripts/preprocess.py /path/to/data /path/to/sensitive_words.txt

TBD

---