MyFriend

This articale provides a detailed overview of the MyFriend model, which has been fine-tuned for specific use cases in emergency situations and environmental issues. The model was developed as part of a hackathon and is designed to assist in generating responses related to these domains. Model Details Model Description

The MyFriend model is a fine-tuned version of the TinyLlama, optimized for generating text in response to queries related to emergency situations and environmental issues. This model was trained on synthetic data generated using the Meta-Llama-3.1-405B-Instruct-Turbo model, with the fine-tuning process conducted on Kaggle using T4 x2 GPUs.

Developed by: Mixed Intelligence Team, led by Umar Majeed
Funded by: Self-funded as part of a hackathon project
Shared by: Umar Majeed
Model type: Text Generation, Conversational AI
Language(s): English
License: Apache 2.0
Finetuned from model: TinyLlama/TinyLlama-1.1B-Chat-v1.0

Model Sources

Repository: MyFriend Model on Hugging Face
Demo: [Link to Demo (if available)]

Uses Direct Use

The MyFriend model is intended to be used directly in applications requiring text generation related to emergency situations and environmental issues. It is suitable for chatbot implementations, emergency response systems, and educational tools focusing on environmental awareness. Downstream Use

The model can be further fine-tuned or integrated into larger systems where specific domain knowledge or custom applications are required. Out-of-Scope Use

The model is not suitable for general-purpose text generation tasks unrelated to its fine-tuned domain. Misuse for generating harmful or misleading information is strongly discouraged. Bias, Risks, and Limitations

The MyFriend model, while fine-tuned on specific data, may still exhibit biases present in the original TinyLlama model. Users should be aware of potential biases, especially in sensitive contexts such as emergency responses. Recommendations

Awareness: Users should be mindful of the model's limitations and biases.
Testing: It is recommended to test the model thoroughly in the intended environment before deployment.

How to Get Started with the Model

To get started with the MyFriend model, use the code snippet below

import torch
from transformers import pipeline

pipe = pipeline("text-generation", model="umarmajeedofficial/MyFriend", torch_dtype=torch.bfloat16, device_map="auto")

messages = [
{
    "role": "system",
    "content": "You are an emergency response assistant with expertise in environmental issues.",
},
{"role": "user", "content": "What should I do during a heat wave?"},
]
prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])

Training Details Training Data

The model was fine-tuned using synthetic data generated from the Meta-Llama-3.1-405B-Instruct-Turbo model. This data was specifically designed to cover a wide range of scenarios related to emergency situations and environmental issues. Training Procedure

Preprocessing: The data was preprocessed to ensure relevance and quality for the fine-tuning task.
Training Hyperparameters: The model was trained using a mixed precision training regime on Kaggle with T4 x2 GPUs.

Evaluation Testing Data, Factors & Metrics Testing Data

Testing was conducted on a subset of the synthetic data generated, with evaluations focusing on the model's ability to provide accurate and contextually appropriate responses in emergency and environmental scenarios. Metrics

Accuracy: The model's ability to generate correct information.
Relevance: The relevance of the generated text to the input query.
Bias Analysis: Evaluation of potential biases in the responses.

Results

The MyFriend model showed strong performance in generating accurate and relevant responses within its fine-tuned domain. Further details on evaluation metrics and results can be found in the repository. Environmental Impact

The environmental impact of training the MyFriend model was minimized by leveraging efficient hardware and cloud resources. The model was trained on Kaggle with T4 x2 GPUs, balancing performance and energy consumption.

Hardware Type: T4 x2 GPUs
Hours used: Approximately 10 hours
Cloud Provider: Kaggle
Compute Region: [Region Information Needed]
Carbon Emitted: Estimated using the Machine Learning Impact calculator

Technical Specifications Model Architecture and Objective

The MyFriend model is built on the TinyLlama architecture, with a focus on conversational text generation. Compute Infrastructure

Hardware: T4 x2 GPUs
Software: The model was fine-tuned using the Hugging Face Transformers library.

Model Card Authors

Umar Majeed (Team Lead) www.linkedin.com/in/umarmajeedofficial
Mixed Intelligence Team Members:
    Moazzan Hassan https://www.linkedin.com/in/moazzan-hassan/
    Shahroz Butt https://www.linkedin.com/in/shahroz-butt-69a813211?utm_source=share&utm_campaign=share_via&utm_content=profile&utm_medium=android_app
    Sidra Hammed https://www.linkedin.com/in/sidra-hameed-8s122000?utm_source=share&utm_campaign=share_via&utm_content=profile&utm_medium=android_app
    Muskan Liaqat https://www.linkedin.com/in/muskan-liaquat-838880308?utm_source=share&utm_campaign=share_via&utm_content=profile&utm_medium=android_app
    Sana Qaisar https://www.linkedin.com/in/sana-qaisar-03b354316/

Model Card Contact

For further information or inquiries, please contact Umar Majeed via Hugging Face.