gemma2b-webxr-showroom-v2/README.md

---
library_name: transformers
pipeline_tag: text-generation
base_model: google/gemma-2b
language:
- en
tags:
- text-generation
- conversational
- small-language-model
- webxr
- virtual-assistant
- xr-ai
- babylonjs
- immersive-ai
datasets:
- custom
license: gemma
---


# gemma2b-webxr-showroom-v2

Fine-tuned Small Language Model designed for **AI-assisted interactions inside WebXR virtual showrooms and immersive product environments**.

This model powers conversational assistants that guide users through **3D product experiences**, explain features, and answer questions in immersive environments.

---

# Model Details

## Model Description

**gemma2b-webxr-showroom-v2** is a fine-tuned conversational model based on **Gemma 2B**.
It is optimized to act as an **AI showroom assistant** in immersive XR applications.

The model was trained to generate responses related to:

* product explanations
* feature descriptions
* interactive showroom guidance
* conversational product queries
* virtual retail assistance

The model is part of the **IntelliShop XR project**, which demonstrates how **AI assistants can enhance WebXR product exploration experiences**.

---

### Developed by

Rajalakshmi Mahadevan (Ramya)

### Model Type

Causal Language Model (Text Generation)

### Language

English

### License

Gemma license (inherits base model licensing requirements)

### Finetuned From

google/gemma-2b

---

# Model Sources

Repository
https://huggingface.co/ramyaa1113/gemma2b-webxr-showroom-v2

Project Context
IntelliShop XR – AI Assisted Virtual Showroom

---

# Intended Uses

## Direct Use

This model is designed to function as a **virtual assistant inside immersive environments**.

Example uses include:

* WebXR virtual product showrooms
* immersive e-commerce experiences
* AI guides inside 3D environments
* product demonstration assistants
* conversational retail bots

Example interaction:

User
Tell me about this XR headset.

Assistant
This XR headset features a high-resolution display, inside-out tracking, and hand tracking support designed for immersive experiences.

---

## Downstream Use

The model can be integrated into larger systems such as:

* WebXR applications
* Babylon.js interactive environments
* AI powered virtual stores
* conversational interfaces for immersive applications

Typical architecture:

WebXR Application
→ Backend API
→ gemma2b-webxr-showroom-v2
→ AI response returned to user

---

## Out-of-Scope Use

This model is **not intended for**:

* medical advice
* legal consultation
* financial decision making
* safety critical systems

The model is optimized specifically for **interactive product assistance scenarios**.

---

# Bias, Risks, and Limitations

Like most language models, this model may:

* produce incorrect or incomplete information
* generate hallucinated details
* reflect biases present in training data

Additionally, the model is **domain tuned**, meaning performance may degrade for topics unrelated to product explanations or showroom interactions.

---

# Recommendations

To improve reliability:

* use structured product metadata as context
* restrict prompts to product related queries
* implement response validation in production systems

---

# How to Use the Model

Example using Transformers:

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "ramyaa1113/gemma2b-webxr-showroom-v2"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = "Explain the features of this XR headset."

inputs = tokenizer(prompt, return_tensors="pt")

outputs = model.generate(
    **inputs,
    max_new_tokens=150
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

---

# Training Details

## Training Data

A custom dataset of approximately **15,000 samples** was created for training.

The dataset includes:

* product explanation prompts
* conversational showroom interactions
* feature descriptions
* assistant style product responses
* retail dialogue examples

The dataset was curated to simulate **real user interactions inside virtual showrooms**.

---

## Training Procedure

### Training Environment

Initial experiments were conducted using **Google Colab**, but runtime instability and GPU limits caused interruptions.

Training was then migrated to **Kaggle GPU notebooks**, which provided a more stable environment for completing the training pipeline.

### Training Regime

Mixed precision training (fp16)

---

# Evaluation

Formal benchmark evaluation has not yet been conducted.
Evaluation currently focuses on **qualitative testing within XR product interaction scenarios**.

Testing scenarios include:

* product explanation quality
* conversational response clarity
* interactive assistant behavior in virtual environments

---

# Environmental Impact

Estimated training environment:

Hardware Type
NVIDIA T4 GPU

Compute Platform
Kaggle Notebooks

Training Duration
Several training runs across multiple sessions

Carbon emissions are estimated to be relatively low due to the small model size and limited training duration.

---

# Technical Specifications

## Model Architecture

Base architecture
Gemma 2B transformer decoder

Task
Causal language modeling (text generation)

---

## Compute Infrastructure

### Hardware

NVIDIA T4 GPU (Kaggle)

### Software

Python
PyTorch
Transformers
Hugging Face ecosystem

---

# Author

Rajalakshmi Mahadevan (Ramya)

XR and AI Developer working at the intersection of:

* Extended Reality (XR)
* WebXR
* Real-time 3D systems
* AI-powered immersive experiences

---

# Model Card Contact

For questions or collaboration:

Hugging Face
https://huggingface.co/ramyaa1113