|
--- |
|
license: apache-2.0 |
|
base_model: internlm/internlm2-chat-7b |
|
datasets: |
|
- CAS-SIAT-XinHai/CPsyCoun |
|
language: |
|
- zh |
|
--- |
|
|
|
# CPsyCounX |
|
|
|
This model is a fine-tuned version of [internlm/internlm2-chat-7b](https://huggingface.co/internlm/internlm2-chat-7b) on the [CPsyCounD](https://huggingface.co/datasets/CAS-SIAT-XinHai/CPsyCoun) dataset. |
|
|
|
## Model description |
|
|
|
**CPsyCounX** is a large language model designed for Chinese Psychological Counseling. |
|
|
|
## Training and evaluation data |
|
|
|
- Train: [CPsyCounD](https://huggingface.co/datasets/CAS-SIAT-XinHai/CPsyCoun) |
|
- Evaluate: [CPsyCounE](https://github.com/CAS-SIAT-XinHai/CPsyCoun/tree/main/CPsyCounE) |
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 1e-06 |
|
- train_batch_size: 4 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- distributed_type: multi-GPU |
|
- num_devices: 4 |
|
- gradient_accumulation_steps: 28 |
|
- total_train_batch_size: 448 |
|
- total_eval_batch_size: 32 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: cosine |
|
- num_epochs: 9.0 |
|
- mixed_precision_training: Native AMP |
|
|
|
### Training results |
|
|
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.37.1 |
|
- Pytorch 2.1.2+cu121 |
|
- Datasets 2.16.1 |
|
- Tokenizers 0.15.1 |
