|
# phi3 |
|
|
|
[![Model Card](https://img.shields.io/badge/Hugging%20Face-Model%20Card-blue)](https://huggingface.co/username/phi3) |
|
|
|
## Description |
|
|
|
**phi3** is a fine-tuned version of phi-3, specifically trained on mental health therapist conversational data. This model is designed to assist in mental health support, providing empathetic and knowledgeable responses in a conversational setting. |
|
|
|
## Installation |
|
|
|
To use this model, you will need to install the following dependencies: |
|
|
|
```bash |
|
pip install transformers |
|
pip install torch # or tensorflow depending on your preference |
|
``` |
|
|
|
## Usage |
|
|
|
Here is how you can load and use the model in your code: |
|
|
|
```python |
|
from transformers import AutoTokenizer, AutoModelForCausalLM |
|
|
|
tokenizer = AutoTokenizer.from_pretrained("username/phi3") |
|
model = AutoModelForCausalLM.from_pretrained("username/phi3") |
|
|
|
# Example usage |
|
chat_template = """ |
|
<|system|> |
|
You are a compassionate mental health therapist. You listen to your clients attentively and provide thoughtful, empathetic responses to help them navigate their emotions and mental health challenges. |
|
<|end|> |
|
<|user|> |
|
I've been feeling really down lately. What should I do? |
|
<|end|> |
|
<|assistant|> |
|
""" |
|
|
|
inputs = tokenizer(chat_template, return_tensors="pt") |
|
outputs = model.generate(**inputs) |
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True) |
|
|
|
print(response) |
|
``` |
|
|
|
### Inference |
|
|
|
Provide example code for performing inference with your model: |
|
|
|
```python |
|
# Example inference |
|
user_input = "I've been feeling really down lately. What should I do?" |
|
chat_template = f""" |
|
<|system|> |
|
You are a compassionate mental health therapist. You listen to your clients attentively and provide thoughtful, empathetic responses to help them navigate their emotions and mental health challenges. |
|
<|end|> |
|
<|user|> |
|
I've been feeling really down lately. What should I do? |
|
<|end|> |
|
<|assistant|> |
|
""" |
|
|
|
inputs = tokenizer(chat_template, return_tensors="pt") |
|
outputs = model.generate(**inputs) |
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True) |
|
|
|
print(response) |
|
``` |
|
|
|
### Training |
|
|
|
If your model can be trained further, provide instructions for training: |
|
|
|
```python |
|
# Example training code |
|
from transformers import Trainer, TrainingArguments |
|
|
|
training_args = TrainingArguments( |
|
output_dir="./results", |
|
evaluation_strategy="epoch", |
|
per_device_train_batch_size=8, |
|
per_device_eval_batch_size=8, |
|
num_train_epochs=3, |
|
weight_decay=0.01, |
|
) |
|
|
|
trainer = Trainer( |
|
model=model, |
|
args=training_args, |
|
train_dataset=train_dataset, |
|
eval_dataset=eval_dataset, |
|
) |
|
|
|
trainer.train() |
|
``` |
|
|
|
## Training Details |
|
|
|
### Training Data |
|
|
|
The model was fine-tuned on a dataset of conversational data from mental health therapy sessions. This dataset includes a variety of scenarios and responses typical of therapeutic interactions to ensure the model provides empathetic and helpful advice. |
|
|
|
### Training Procedure |
|
|
|
The model was fine-tuned using a standard training approach, optimizing for empathy and relevance in responses. Training was conducted on [describe hardware, e.g., GPUs, TPUs] over [number of epochs] epochs with [any relevant hyperparameters]. |
|
|
|
## Evaluation |
|
|
|
### Metrics |
|
|
|
The model was evaluated using the following metrics: |
|
|
|
- **Accuracy**: X% |
|
- **Empathy Score**: Y% |
|
- **Relevance Score**: Z% |
|
|
|
### Comparison |
|
|
|
The performance of phi3 was benchmarked against other conversational models in the mental health domain, demonstrating superior empathy and contextual understanding. |
|
|
|
## Limitations and Biases |
|
|
|
While phi3 is highly effective, it may have limitations in the following areas: |
|
- It may not be suitable for providing critical mental health interventions. |
|
- There may be biases present in the training data that could affect responses. |
|
|
|
## How to Contribute |
|
|
|
We welcome contributions! Please see our [contributing guidelines](link_to_contributing_guidelines) for more information on how to contribute to this project. |
|
|
|
## License |
|
|
|
This model is licensed under the [MIT License](LICENSE). |
|
|
|
## Acknowledgements |
|
|
|
We would like to thank the contributors and the creators of the datasets used for training this model. |
|
``` |
|
|
|
### Tips for Completing the Template |
|
|
|
1. **Replace placeholders** (like `username`, `training data`, `evaluation metrics`) with your actual data. |
|
2. **Include any additional information** specific to your model or training process. |
|
3. **Keep the document updated** as the model evolves or more information becomes available. |