README.md

metadata

license: mit
language:
  - en
base_model:
  - google/gemma-1.3b-it
tags:
  - empathy
  - emotion
  - chatbot
  - feeling
  - friendly-ai
pipeline_tag: text-generation

🧸 Empathy Chatbot — Fine-tuned GEMMA for Emotional Conversations

Model ID: sajeewa/empathy-chat-gemma
This is a fine-tuned version of google/gemma-1.3b-it designed to respond with care, warmth, and empathy in emotional conversations. It's trained on the EmpatheticDialogues dataset to make it emotionally aware and conversationally comforting — like a caring friend who calls you “baby” or “cutey” and sprinkles in sweet emojis 🧸💖.

---

🧠 Model Details

• Base model: google/gemma-1.3b-it
• Fine-tuned with: Unsloth + 🤗 TRL
• Dataset: EmpatheticDialogues
• Training location: Kaggle (2×T4 GPUs)
• Intended use: Friendly, emotionally supportive chatbots

---

💬 Chat Template & Interface

This model uses Hugging Face’s chat template format. The chatbot behaves like a sweet and caring friend who responds with emotionally intelligent and supportive language, using cute nicknames and emojis. Here's how you can interact with it:

from transformers import AutoTokenizer, AutoModelForCausalLM, TextStreamer
import torch

model_id = "sajeewa/empathy-chat-gemma"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto"
)

chat_history = [
    {
        "role": "system",
        "content": (
            "You are an empathetic AI and your friend. Always give lovely caring messages. "
            "Understand the user's feelings. Then provide a caring response. "
            "Please give responses as a good friend, using lovely words like 'baby', 'my cutey', etc. 💖 "
            "Use emojis to be calming 😊. Continue conversations with a warm tone."
        )
    }
]

user_input = "I'm feeling lonely today."
chat_history.append({"role": "user", "content": user_input})

prompt = tokenizer.apply_chat_template(
    chat_history,
    tokenize=False,
    add_generation_prompt=True,
)

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

output = model.generate(
    **inputs,
    max_new_tokens=128,
    temperature=0.7,
    top_p=0.95,
    top_k=50,
    do_sample=True,
    streamer=streamer
)

response = tokenizer.decode(output[0], skip_special_tokens=True)
print(response)