Model Download

WeClone uses the Qwen2.5-VL-7B model by default, but you can choose other models you want to use (thinking models are not recommended).

TIP

7B models generally have poor results (although some feedback suggests they can be acceptable, which may be related to the data), 14B is barely acceptable, and 32B is better.

If you have chosen another model or downloaded the model to a different directory, please modify it in settings.jsonc:

"common_args": {
    "model_name_or_path": "your_model_path",
    "template": "your_model_template",
    ...
},
"train_sft_args": {
    "lora_target": "module_names_of_the_model_to_be_fine-tuned", // e.g., "q_proj", "k_proj", "v_proj", "o_proj"
    ...
},

Method 1: Download using the Hugging Face command-line tool

huggingface-cli download Qwen/Qwen2.5-7B-Instruct --local-dir ./models/Qwen2.5-7B-Instruct --local-dir-use-symlinks False

⚠️ The model file is large, and the download may take a long time. Please ensure you have sufficient disk space (at least 20GB is recommended).
The download process is long, and the connection may be interrupted. You can resume with an incremental download.

Method 2: Install using Git

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