---
base_model: openai-community/gpt2-medium
pipeline_tag: question-answering
license: mit
datasets:
- SohamGhadge/casual-conversation
language:
- en
metrics:
- accuracy
library_name: peft
---
## 🧠 Fine-Tuned GPT-2 Medium for Conversational AI

This project fine-tunes the `gpt2-medium` language model to support natural, casual **conversational dialogue** using **PEFT + LoRA**.

---

### 🚀 Model Summary

* **Base model**: `gpt2-medium`
* **Objective**: Enable natural question-answering and dialogue
* **Training method**: Supervised Fine-Tuning (SFT) using PEFT with LoRA adapters
* **Tokenizer**: `gpt2` (same as base model)

---

### 📈 Training Metrics

| Metric              | Value          |
| ------------------- | -------------- |
| Global Steps        | 2611           |
| Final Training Loss | 2.185          |
| Training Runtime    | 430.61 seconds |
| Samples/sec         | 138.41         |
| Steps/sec           | 17.32          |
| Total FLOPs         | 1.12 × 10¹⁵    |
| Epochs              | 7.0            |

> These metrics reflect final performance after complete training.

---

### 💬 Inference Script

Chat with the model using the `talk()` function below:

```python
def talk(model=peft_model, tokenizer=tokenizer, device=device):
    print("Start chatting with the bot! Type 'exit' to stop.\n")
    while True:
        question = input("You: ")
        if question.lower() == "exit":
            print("Goodbye!")
            break

        prompt = f"User: {question}\nBot:"
        inputs = tokenizer(prompt, return_tensors="pt").to(device)

        with torch.no_grad():
            outputs = model.generate(
                **inputs,
                max_new_tokens=20,
                do_sample=True,
                temperature=0.7,
                top_p=0.9,
                pad_token_id=tokenizer.eos_token_id
            )

        response = tokenizer.decode(
            outputs[0][inputs["input_ids"].shape[-1]:],
            skip_special_tokens=True
        )

        # Clean response
        response = response.split(".")
        response = ".".join(response[:-1]) + "."
        print("Bot:", response.strip())
```

* 🤖 **Stateless**: No memory across turns (yet).
* 🌱 **Future idea**: Add memory/context for multi-turn dialogue.

---

### ⚙️ Quick Setup

To use this model locally:

```bash
pip install transformers peft accelerate
```

---