Gemma2 Korean Emotional Dialogue Chatbot

image/png

https://www.kaggle.com/code/hwkims/gemma2-for-korean-mental-wellbeing-inference-note/notebook

https://www.kaggle.com/code/hwkims/fork-of-gemma2-for-korean-mental-wellbeing-infere/notebook?scriptVersionId=221718992

https://huggingface.co/spaces/kimhyunwoo/gemma2-korean-emotion-chat

This Space demonstrates a fine-tuned Gemma 2 2B model for generating empathetic Korean dialogue. The model is trained on the AI Hub "Emotional Dialogue Corpus" using Low-Rank Adaptation (LoRA) and is designed to respond to user prompts that express various emotions.

Model Details

  • Model: Gemma 2 2B
  • Fine-tuning Method: Low-Rank Adaptation (LoRA)
  • Training Data: AI Hub "Emotional Dialogue Corpus"
  • Language: Korean
  • Task: Emotional Dialogue Generation

How to Use

  1. Enter your Korean text in the "User Input" text box.
  2. Optionally, you can adjust parameters such as "System message", "Max new tokens", "Temperature", and "Top-p" for diverse response generation.
  3. Click "Submit" to get the chatbot's response.
  4. The model's output will appear in the "Gemma Output" text box.

Example Prompts

  • 였늘 λ„ˆλ¬΄ νž˜λ“€μ–΄μš”. (I'm so tired today.)
  • 기뢄이 λ„ˆλ¬΄ μ’‹μ•„μš”! (I'm in a great mood!)
  • κ±±μ •λ˜λŠ” 일이 μžˆμ–΄μ„œ 잠이 μ•ˆ μ™€μš”. (I can't sleep because I'm worried.)

Limitations

  • The model might not fully understand complex nuances, sarcasm, or irony.
  • The model's response can be influenced by the given system message.
  • The model's output is not a substitute for professional mental health support.

Code and Resources

The code for this Space is available in the "Files" tab.

About

This Space was created as a part of the Google - Unlock Global Communication with Gemma competition. It showcases the fine-tuning of Gemma 2 for Korean emotional dialogue, aiming to provide culturally sensitive and empathetic support.

License

This Space and the underlying model are licensed under the Gemma license.

Downloads last month
80
Safetensors
Model size
2.62B params
Tensor type
FP16
Β·
Inference Providers NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API: The model has no pipeline_tag.

Model tree for kimhyunwoo/gemma2-ko-dialogue-lora

Base model

google/gemma-2-2b
Finetuned
(530)
this model