Loading repository data…
whyzhow / repository
Let AI think and express like you. This framework provides a complete assembly line: from the noise processing of original chat records, to the seamless switching of multi-model adaptation layers, to local lightweight fine-tuning (LoRA).
Let AI think and express like you. This framework provides a complete assembly line: from the noise processing of original chat records, to the seamless switching of multi-model adaptation layers, to local lightweight fine-tuning (LoRA).
This project divides the personalisation process of AI into three core stages:
Convert messy chat records (including emojis, links, spam) into high-quality conversation pairs.
Through a unified interface, your personal style can be easily loaded to GPT-4, Claude or local Llama 3.
Use fine-tuning technology to solidify your language habits in the weight of the model, not just rely on Prompt.
Input (original chat.txt):
User A: Have you eaten yet?
I: Eat, [expression] I'm really hungry, https://link.com
User A: Then let's go.
Me: Indeed, Let'S Go.
Output (JSONL after cleaning):
{
"Messages": [
{"Role": "user", "content": "Have you eaten yet?"},
{"Role": "assistant", "content": "I'm really hungry after eating"}
]
}
PersonalStyleAI-Framework/
├── data/
│ ├── raw/ # Original chat records (such as chat.txt)
│ └── processed/ # JSONL training data set after cleaning
├── src/ # Source code
│ ├── core/ # adaptor logic and factory mode implementation
│ ├── utils/ # Data preprocessing and string cleaning tools
│ └── trainers/ # Model fine-tuning script (based on PEFT/LoRA)
├── pyproject.toml # Modern Python Dependency and Project Configuration
├── preprocess_data.py # Data processing entry script
├── main.py # Style Dialogue Test Entrance
└── .env.example # Environment variable template
Adopt factory model design. This means that if you want to switch from OpenAI to local Ollama, you only need to change the one-line configuration without rewriting the business logic.
Efficient regular expressions are preset and optimised for text exported by social software.
Use .env to manage sensitive information and manage dependency hierarchy through pyproject.toml.
1.Basic version (only call API)
# Clone Project
Git clone [https://github.com/your username/PersonalStyleAI-Framework.git](https://github.com/your username/Personal StyleAI-Framework.git)
Cd PersonalStyleAI-Framework
# Create a virtual environment and install core dependencies
Python -m venv venv
Source venv/bin/activate # Windows use venv\Scripts\activate
Pip install-e.
Create a .env file and fill in your API Key:
cp .env.example .env
Collect data: Put your chat records or articles into data/raw/chat.txt. Running cleaning:
python preprocess_data.py
The script will generate data/processed/train.jsonl, which is a "textbook" for AI to learn your style. 4. Run the dialogue
python main.py
If you have a graphics card that supports CUDA, you can install fine-tuning components for local training: Pip install -e ".[train]"
# Run the fine-tuning script (need to configure parameters according to src/trainers)
Python run_train.py
If you have any suggestions for improvement or want to add more AI adaptors (such as Anthropic or DeepSeek), welcome to submit a Pull Request or open an Issue discussion.
Selected from shared topics, language and repository description—not editorial ratings.
kontentbuzdygan /
Hello beautiful people, here is wideobuzdygan from kontentbuzdygan just for you. Let us know what you think! Greeting greetings