Skip to content

Latest commit

 

History

History
45 lines (34 loc) · 1.61 KB

README.md

File metadata and controls

45 lines (34 loc) · 1.61 KB

Qwen2 -> Named Entity Recognition

Note

With the latest update, the pipeline now supports high-level training control through a YAML file, eliminating the need to modify source code, except when adding a new dataset. For new datasets, you must still convert your dataset to the required format. For all other cases, simply modify the configuration file (i.e., train_config.yaml) to make necessary adjustments.

Basic Information

Setup Details

  • Accelerator: NVIDIA RTX 4090D $\times$ 2
  • Platform: Linux
  • Internet: Enabled

Model and Resources

Additional Information

Important

Key modules are implemented in the qwen2ner module. For more technical details, please refer to the module.

  • Download the dataset from Kaggle to the dataset folder.

  • Construct the .csv format dataset.

python3 construct_text_data.py
  • Train the model.
./train.sh
  • Inference on a single text.
python3 inference.py \
    --model_name_or_path MODEL_NAME_OR_PATH

Blogs / 中文博客

  • 知乎：【大模型微调】Qwen SFT：基于 trl 框架的 QLoRA 微调
  • CSDN：【大模型微调】Qwen SFT：基于 trl 框架的 QLoRA 微调