xming521/WeClone

简体中文｜ English｜ Project Homepage ｜ Documentation

• 💫 Complete end-to-end solution for creating digital avatars, including chat data export, preprocessing, model training, and deployment
• 💬 Fine-tune LLM using chat history with support for image modal data, infusing it with that authentic "flavor"
• 🔗 Integrate with Telegram, WhatsApp (coming soon) to create your own digital avatar
• 🛡️ Privacy information filtering with localized fine-tuning and deployment for secure and controllable data

感谢 Infistar.cc 无限星河 赞助本项目！
⚡ 稳定高效的模型服务：价格低至官方渠道 1 折，模型倍率公开透明，多节点冗余保障，降低限流、429 与服务中断的影响；
🧠 主流模型一站式接入：一个 API Key 即可调用 Claude、ChatGPT、Gemini、Kimi、GLM、DeepSeek 等模型；
👤 助力数字分身生态扩展：适用于对话效果测试、人格提示词优化、内容生成及机器人能力扩展，为 WeClone 数字分身提供更多模型选择；
🎁 WeClone 用户专属福利：通过 专属推广链接 注册即可领取 5 美元等值测试额度 / 首充专属优惠，快速体验多模型 API 服务！

[25/07/10] Data source added Telegram
[25/06/05] Support for image modal data fine-tuning

The project uses Qwen2.5-VL-7B-Instruct model by default with LoRA method for SFT stage fine-tuning. You can also use other models and methods supported by LLaMA Factory.

Estimated VRAM requirements:

1. CUDA installation (skip if already installed, requires version 12.6 or above)

2. It is recommended to use uv to install dependencies, which is a very fast Python environment manager. After installing uv, you can use the following commands to create a new Python environment and install dependencies.

git clone https://github.com/xming521/WeClone.git && cd WeClone
uv venv .venv --python=3.12
source .venv/bin/activate # windows .venv\Scripts\activate
uv pip install --group main -e . 

1. Copy the configuration file template and rename it to settings.jsonc, and make subsequent configuration changes in this file:

cp examples/tg.template.jsonc settings.jsonc

1. Use the following command to test whether the CUDA environment is correctly configured and can be recognized by PyTorch (not needed for Mac):

  python -c "import torch; print('CUDA Available:', torch.cuda.is_available());"

1. (Optional) Install FlashAttention to accelerate training and inference: uv pip install flash-attn --no-build-isolation.

It is recommended to use Hugging Face to download models, or use the following command:

git lfs install
git clone https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct models/Qwen2.5-VL-7B-Instruct

Please use Telegram Desktop to export chat records. Click the top right corner in the chat interface, then click "Export chat history". Select Photos for message types and JSON for format. You can export multiple contacts (group chat records are not recommended), then place the exported ChatExport_* in the ./dataset/telegram directory, meaning put different people's chat record folders together in ./dataset/telegram.

• First, modify the language, platform, and include_type in the configuration file according to your needs.
• If you use telegram, you need to modify the telegram_args.my_id in the configuration file to your own telegram user ID.
• By default, the project uses Microsoft Presidio to remove phone numbers, email addresses, credit card numbers, IP addresses, geographic location names, international bank account numbers, cryptocurrency wallet addresses, age information, and generic ID numbers from the data, but it cannot guarantee 100% identification.
• Therefore, a blocklist blocked_words is provided in settings.jsonc, allowing users to manually add words or phrases they want to filter (the entire sentence containing blocked words will be removed by default).

• Execute the following command to process the data. You can modify the make_dataset_args in settings.jsonc according to your own chat style.

weclone-cli make-dataset

More Parameter Details: Data Preprocessing

Run from the project root:

weclone-cli build-profile

Choose profiles only (the default) or include events. The results may overlap; choose profiles only to save tokens. You can also select the mode directly:

weclone-cli build-profile --profile-only
weclone-cli build-profile --with-events

Results are saved under dataset/res_csv/agent/profile_runs/, and the profile path is printed on completion. Use --output-dir <new-directory> to choose the result directory. API models require --max-context-tokens <limit> to set the input token limit.

Run weclone-cli server to view and review profiles, or weclone-cli server --source <profile-path> to select a profile.

Set security_args in settings.jsonc:

"security_args": {
    "storage_mode": "encrypted"
}

New templates default to encrypted; choose plaintext for plaintext storage. Older configurations without this setting use plaintext. Encrypted mode does not yet support distributed training.

Set a shared password at the first encrypted task or web visit. Use it to unlock later tasks and sign in to the web interface.

Encrypted outputs end in .enc. Protect your original chats and any plaintext exports. Keep the .weclone/ directory: it contains the key needed to read encrypted data.

Export plaintext, change a password, or reset a forgotten password:

weclone-cli decrypt --input <encrypted-file> --output <new-plaintext-file>
weclone-cli security-change-password  # Requires the old password; preserves data
weclone-cli server-reset-password     # Reset a forgotten password

Encrypted data cannot be recovered without the password. After resetting it or switching storage modes, regenerate data from the original chats in a new output directory.

• (Optional) Modify model_name_or_path, template, lora_target in settings.jsonc to select other locally downloaded models.
• Modify per_device_train_batch_size and gradient_accumulation_steps to adjust VRAM usage.
• You can modify parameters like num_train_epochs, lora_rank, lora_dropout in train_sft_args based on your dataset's quantity and quality.

weclone-cli train-sft

Uncomment the deepspeed line in settings.jsonc and use the following command for multi-GPU training:

uv pip install "deepspeed<=0.16.9"
deepspeed --num_gpus=number_of_gpus weclone/train/train_sft.py

Test suitable temperature and top_p values, then modify infer_args in settings.jsonc for subsequent inference use.

weclone-cli webchat-demo

weclone-cli server --inference --port 8005

Does not include questions asking for personal information, only daily conversation. Test results are in test_result-my.txt.

weclone-cli server --inference --port 8005
weclone-cli test-model

AstrBot is an easy-to-use multi-platform LLM chatbot and development framework ✨ Supports Discord, Telegram, Slack, Feishu and other platforms.

Usage steps:

1. Deploy AstrBot
2. Deploy messaging platforms like Discord, Telegram, Slack in AstrBot
3. Execute weclone-cli server --inference --port 8005 to start the API service
4. Add a new service provider in AstrBot, select OpenAI type, fill in the API Base URL according to AstrBot's deployment method (e.g., for docker deployment it might be http://172.17.0.1:8005/v1), fill in the model as gpt-3.5-turbo, and enter any API Key
5. Tool calling is not supported after fine-tuning, please turn off the default tools first by sending the command: /tool off_all on the messaging platform, otherwise the fine-tuned effect won't be visible.
6. Set the system prompt in AstrBot according to the default_system used during fine-tuning.

LangBot is an easy-to-use open-source LLM chatbot platform suitable for various scenarios. It connects to various global instant messaging platforms. You can set up your IM bot in just 5 minutes.

1. Deploy LangBot
2. Add a bot (Discord, Telegram, Slack, Lark e.g.) in LangBot
3. Execute weclone-cli server --inference --port 8005 to start the WeClone API service
4. Add a new model in the model page, name it gpt-3.5-turbo, select OpenAI as the provider, fill in the request URL as WeClone's address. For detailed connection methods, refer to the documentation, and enter any API Key.

1. Select the model you just added in the pipeline configuration, or modify the prompt configuration

• Support more data sources
• Richer context: including contextual conversations, chat participant information, time, etc.
• Memory support
• Multimodal support: image support already implemented
• Data augmentation
• GUI support
• COT (Chain of Thought) thinking support

It is also recommended to use DeepWiki for problem solving.

Any Issues/Pull Requests are welcome!

You can contribute by checking Issues or helping review PRs (Pull Requests). For new feature additions, please discuss through Issues first.
Development environment:

uv pip install --group dev -e .
pre-commit install

The project uses pytest for testing, pyright for type checking, and ruff for code formatting.
Before submitting your code, you should run pytest tests to ensure all tests pass.

Thanks to the following code contributors and other community members for their contributions

This project also benefits from excellent open source projects such as PyWxDump, LLaMA-Factory, AstrBot, LangBot, and others.

Please carefully read and understand all contents of this disclaimer, ensuring strict compliance with relevant regulations when using this project.