---
base_model:
- google/gemma-2-2b-it
library_name: transformers
license: apache-2.0
language:
- sw
- en
metrics:
- accuracy
- bleu
pipeline_tag: text2text-generation
datasets:
- fka/awesome-chatgpt-prompts
tags:
- finance
- AI
- NLP
- customer-support
new_version: google/gemma-2-27b-it
---

# Model Card for Futuresony AI

## Model Details

### Model Description
Futuresony AI is a fine-tuned language model designed to enhance conversational AI in customer support, finance, and general-purpose NLP applications. It is developed with a focus on multilingual capabilities, supporting both Swahili and English. The model is particularly fine-tuned for tasks such as text generation, question-answering, and AI-driven customer support solutions, aligning with Salum A. Salum’s goal of introducing AI-driven automation in Tanzania.

- **Developed by:** Futuresony Tech
- **Funded by:** Futuresony
- **Shared by:** Salum A. Salum
- **Model type:** Transformer-based text-to-text generation model
- **Language(s) (NLP):** Swahili and English
- **License:** Apache 2.0
- **Finetuned from model:** google/gemma-2-2b-it

### Model Sources
- **Repository:** [More Information Needed]
- **Paper:** [More Information Needed]
- **Demo:** [More Information Needed]

## Uses

### Direct Use
The model can be used for:
- AI-powered customer support for businesses
- Financial query handling
- General NLP tasks like summarization, translation, and chatbot applications
- Integration with WhatsApp, SMS, and social media for automated responses

### Downstream Use
When fine-tuned, the model can be applied to:
- ASA Microfinance, model finetuned with ASA Microfinance Company to assist internal and external users related to ASA questions
- AI-driven traffic control and safety enhancements at Tanga Port
- AI-driven FAQ systems for businesses

### Out-of-Scope Use
- Any application that violates ethical AI usage, such as misinformation, hate speech, or unethical automation.
- Legal or medical advisory applications without proper expert supervision.

## Bias, Risks, and Limitations
While Futuresony AI is trained to provide accurate and helpful responses, it may still exhibit biases present in the training data. There is also a risk of incorrect or misleading responses, particularly for sensitive topics.

### Recommendations
Users should verify important responses with human oversight, especially in critical domains like finance and customer support. Continuous monitoring and retraining are recommended to improve accuracy and fairness.

## How to Get Started with the Model
To use the model, follow this example:
```python
from transformers import pipeline

generator = pipeline("text2text-generation", model="Futuresony/future_ai_12_10_2024.gguf")
response = generator("What is the best way to improve AI chatbot accuracy?")
print(response)
```

## Training Details

### Training Data
The model is fine-tuned using a dataset that includes:
- Customer service inquiries and responses
- Financial support queries
- Swahili-English multilingual conversations

### Training Procedure
- **Preprocessing:** Tokenization and normalization of data
- **Training Regime:** Mixed precision (fp16) for efficiency
- **Hardware:** Trained using Google Colab with TPU acceleration

## Evaluation

### Testing Data, Factors & Metrics

#### Testing Data
- Datasets used: fka/awesome-chatgpt-prompts
- Additional customer service data from Futuresony’s internal datasets

#### Factors
- Response accuracy
- Coherence in multilingual interactions
- Efficiency in real-time conversations

#### Metrics
- **Accuracy:** Measures the correctness of generated responses
- **BLEU Score:** Evaluates text coherence and fluency

### Results
Initial tests show strong performance in customer service scenarios, with high coherence and accuracy in answering user queries in both Swahili and English.

## Model Examination
Further research is needed to evaluate:
- Bias mitigation
- Performance in domain-specific use cases

## Environmental Impact
Training energy consumption:
- **Hardware Type:** Google Colab TPU
- **Hours used:** ~50 hours
- **Cloud Provider:** Google Cloud
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]

## Technical Specifications
### Model Architecture and Objective
Futuresony AI is based on the Google Gemma-2 architecture and fine-tuned for NLP tasks with a focus on customer support and finance applications.

### Compute Infrastructure
- **Hardware:** TPU-based training
- **Software:** Hugging Face Transformers, PyTorch

## Citation
If you use Futuresony AI in your research or project, please cite:
```bibtex
@misc{FuturesonyAI2025,
  author = {Salum A. Salum},
  title = {Futuresony AI: A Multilingual Conversational Model},
  year = {2025},
  url = {https://huggingface.co/Futuresony}
}
```

## Model Card Contact
For inquiries or collaboration, contact:
- **Email:** ally.salum.salum6@gmail.com
- **Phone:** +255 672 087 616

---
This model card provides an overview of Futuresony AI, detailing its purpose, training, and usage guidelines. The information is subject to updates as the project evolves.