GPT2 Student Advisor

Model Overview

Meet the GPT2 Student Advisor, your personal AI counselor who’s been fine-tuned on student data to dole out academic advice. This model is based on GPT-2 and has been trained to generate personalized suggestions for students. Think of it as a non-judgmental guide to tell you (or your students) that maybe skipping classes and not studying isn’t the best strategy. It looks at stuff like study hours, attendance, sleep patterns, and other school-related activities to give advice, whether it’s to buckle down, sleep more, or, in some cases, keep up the good work.

Model Architecture

  • Base model: GPT-2 (because GPT-3 was busy)
  • Fine-tuned: Yep! Specifically for generating suggestions based on student profiles.

What’s This For?

This isn’t just some random text generator. The GPT2 Student Advisor can actually be used to give academic guidance, helping students (or teachers) with actionable advice. It’s especially handy for:

  • Automated Student Advising: Use it as a chatbot to gently nudge students in the right direction.
  • Educational Platforms: Got an education app? Plug this in for a tailored learning experience.
  • Counselor’s Best Friend: Helping counselors generate quick suggestions (but seriously, humans are still important).

Training Data

The model was trained on the Student Performance Factors dataset, which has data on things like:

  • Study Hours
  • Attendance
  • Parental Involvement (aka “Do your parents care about your grades?”)
  • Sleep Hours (yes, sleep matters)
  • Internet Access (because no internet is the modern tragedy)
  • Plus a bunch of other factors that make students tick.

Using this data, the model learned to generate suggestions like "Maybe study more than 5 hours a week" or "Get some sleep." All the good stuff.

Sample Input:

Student Profile:
- Hours Studied per week: 5
- Attendance: 60%
- Parental Involvement: Low
- Access to Resources: Medium
- Extracurricular Activities: None
- Sleep Hours per night: 6
- Previous Scores: 70
- Motivation Level: Low
- Internet Access: None
- Tutoring Sessions per month: 0
- Family Income: Low
- Teacher Quality: Medium
- School Type: Public
- Peer Influence: Negative
- Physical Activity per week: 1 hour
- Learning Disabilities: Yes
- Parental Education Level: High School
- Distance from Home: Far
- Gender: Male

Sample Output:

Suggestions:
- Consider increasing your study hours.
- Improve your class attendance.
- Seek more support from your parents.
- Ensure you get enough sleep each night.
- Find ways to boost your motivation.
- Find ways to access the internet for study resources.
- Consider seeking help for your learning disabilities.
- Engage in more physical activities for better health.

How We Trained It

  • Batch size: 8 (because 16 was too easy)
  • Epochs: 3 (the magic number)
  • Learning rate: 5e-5
  • Optimizer: Good ol’ AdamW
  • Evaluation: After each epoch, because we like checking in on progress.
  • Mixed precision: Enabled, because who doesn't like faster training?

Training Environment

  • Hardware: Powered by an NVIDIA GPU (because anything else just wouldn't cut it).
  • Software: transformers library from Hugging Face, running on PyTorch.

Performance

The model was evaluated using:

  • Loss: Yeah, loss matters here too. We minimized it using causal language modeling.
  • Best Model Selection: After every epoch, we kept the one that made the least mistakes. Low loss = good.

Things to Keep in Mind

  • It's Specialized: This model was trained on student data. Don’t expect it to give life advice outside the academic sphere—though it might be fun to ask!
  • Be Specific: Keep the student profile format consistent for the best results. Random input = random output.

Ethical Stuff

Let’s not pretend this model is a replacement for real counseling. It’s trained on a dataset that makes a lot of assumptions about student behavior and performance. Real life is, of course, more complex. Please use this model as an assistant, not a replacement for human guidance.

How to Use It

from transformers import GPT2LMHeadModel, GPT2Tokenizer

# Load the fine-tuned model and tokenizer
model = GPT2LMHeadModel.from_pretrained("LyubomirT/gpt2-student-suggester")
tokenizer = GPT2Tokenizer.from_pretrained("LyubomirT/gpt2-student-suggester")

# Define a student profile
student_profile = """
Student Profile:
- Hours Studied per week: 5
- Attendance: 60%
- Parental Involvement: Low
- Sleep Hours per night: 6
- Motivation Level: Low
"""

# Tokenize the input and generate suggestions
inputs = tokenizer.encode(student_profile, return_tensors="pt")
outputs = model.generate(inputs, max_length=600, num_beams=5, early_stopping=True)
suggestions = tokenizer.decode(outputs[0], skip_special_tokens=True)

print("Generated Suggestions:")
print(suggestions)

License

This model is released under the MIT license, because we believe in sharing the love. Check out Hugging Face's Model Licensing guidelines for more info.

Downloads last month
19
Safetensors
Model size
124M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for LyubomirT/gpt2-student-suggester

Finetuned
(1239)
this model