You can deploy this model and start earning money today!

mcgowee/
empathetic-t5

Name: empathetic-t5
Availability: InStock

safetensors

Views

No views yet

Empathetic-T5

Model type: t5-small fine-tuned for empathetic response generation Repository: https://huggingface.co/mcgowee/empathetic-t5 License: Apache-2.0

Model Overview

Empathetic-T5 is a sequence-to-sequence model based on t5-small, fine-tuned to generate contextually appropriate empathetic replies given user input and an emotion label. fileciteturn1file2

Dataset

Source dataset: empathetic_dialogues_v2 from Hugging Face (Adapting et al., 2023) fileciteturn1file2
Source fields:
- chat_history: user utterances split into individual prompts
- emotion: emotion label associated with each prompt
Target field:
- system_response: the coach-like empathetic reply

Training Details

Model architecture: t5-small (Text-to-Text Transfer Transformer) fileciteturn1file2
Training script: train_t5.py
Hyperparameters:
- num_train_epochs: 3
- eval_strategy: "epoch"
Libraries:
- Hugging Face transformers
- numpy
- PyTorch
Preprocessing:
- split_chat_into_pairs to break down multi-turn dialogues into (input, response) pairs
Training process: Model was trained on local or rented GPU; limited hyperparameter tuning was performed due to time constraints fileciteturn1file2

Usage Example

python

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

tokenizer = AutoTokenizer.from_pretrained("mcgowee/empathetic-t5")
model = AutoModelForSeq2SeqLM.from_pretrained("mcgowee/empathetic-t5")
empathic_pipeline = pipeline(
    "text2text-generation",
    model=model,
    tokenizer=tokenizer,
    device=0  # use -1 for CPU
)

prompt = "I just had a terrible day and nobody seems to care."
response = empathic_pipeline(prompt, max_length=64)
print(response[0]['generated_text'])

Intended Uses and Limitations

Intended uses:

Generating empathetic replies for chatbots and conversational agents
Demonstrating fine-tuning of a T5 model on empathetic dialogue data

Limitations & Risks:

Not a substitute for professional mental health support
Limited by the quality and diversity of the training data; may produce biased or inappropriate outputs
Additional evaluation and fine-tuning recommended before deployment in sensitive applications

Citation

bibtex

@misc{mcgowee_empathetic-t5_2025,
  title = {Empathetic-T5: T5 Model Fine-Tuned for Empathetic Response Generation},
  author = {Earl McGowen (mcgowee)},
  year = {2025},
  howpublished = {\\url{https://huggingface.co/mcgowee/empathetic-t5}}
}

References

Adapting et al. (2023). empathetic_dialogues_v2 [Data set]. Hugging Face. fileciteturn1file2
Hugging Face. (2024). T5 model documentation. Retrieved from https://huggingface.co/docs/transformers/model_doc/t5