Gemma 2B IT - Customer Support Fine-tuned Model (QLoRA)
This model is a fine-tuned version of google/gemma-1.1-2b-it
using QLoRA on a custom instruction-tuning dataset designed for automating customer support tasks, including:
✉️ Complaint summarization
💬 Sentiment analysis
🧠 Topic modeling
📉 Churn prediction
🧾 Auto response drafting
🛠 Fine-tuning Details
Technique: QLoRA (4-bit quantization using bitsandbytes)
Dataset: 14k+ records combining Amazon review and Q&A data
Data format: ChatML-style JSONL with messages: [{role: ..., content: ...}]
Training platform: Google Colab (A100 GPU)
Libraries: Hugging Face transformers, peft, datasets
💡 Use Cases
This model is best suited for:
Automating customer support replies using auto-response drafting
Summarizing customer complaints to understand the needs better
Classifying topics and customer sentiments
Predicting churn based on interaction tone and topics
