Virtual Professor AI is designed to be used as an educational assistant that can teach any subject. It can answer questions, explain concepts step by step, help with homework, generate quizzes, and provide clear explanations across all academic subjects including math, science, history, programming, literature, and more.

Downstream Use

Educational platforms
Student learning assistants
Online tutoring applications
Self-study tools

Out-of-Scope Use

Medical diagnosis
Legal advice
Financial advice
Generating harmful or misleading content
Replacing certified professional educators

Bias, Risks, and Limitations

The model is based on TinyLlama 1.1B which is a small model and may produce incorrect answers
Training data (Alpaca) may contain biases
Limited to English language
May struggle with very complex or specialized topics
Should not be used as the sole source of information for critical decisions

Recommendations

Users should verify important information from authoritative sources. This model is best used as a learning aid and study companion, not as a replacement for certified educators or textbooks.

How to Get Started with the Model

Using Python

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch

base_model = AutoModelForCausalLM.from_pretrained(
    "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
    torch_dtype=torch.float16,
    device_map="auto"
)

model = PeftModel.from_pretrained(
    base_model, 
    "christianmalonga/virtual-professor"
)

tokenizer = AutoTokenizer.from_pretrained(
    "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
)

prompt = """### Instruction:
What is photosynthesis?

### Response:
"""

inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training Details

Training Data

Dataset: yahma/alpaca-cleaned
Size: 52,000 examples (subset)
Content: Instruction-response pairs covering all academic subjects

Training Procedure

Preprocessing

Dataset formatted into instruction-response format:

### Instruction:
{question}

### Response:
{answer}

Training Hyperparameters

Training regime: bf16 mixed precision
Learning rate: 2e-4
Epochs: 3
Batch size: 4 (per device)
Gradient accumulation steps: 2
Warmup steps: 50
Optimizer: AdamW
LoRA rank (r): 8
LoRA alpha: 16
LoRA dropout: 0.05
Target modules: q_proj, v_proj, k_proj, o_proj

Speeds, Sizes, Times

Training time: ~8 hours 4 minutes
Final loss: 0.077
Model size: 1.2 GB (GGUF Q8_0)
LoRA adapter size: 4.53 MB

Evaluation

Testing Data

Manual evaluation on held-out questions across multiple subjects.

Metrics

Final Training Loss: 0.077
Initial Loss: 1.377
Loss Reduction: 94%

Results

The model successfully answers questions across multiple subjects including:

✅ Science (Photosynthesis, Biology)
✅ Mathematics (Pythagorean theorem)
✅ History (World War 1 causes)
✅ General knowledge

Summary

Virtual Professor AI demonstrates strong instruction following and educational response generation after fine-tuning on the Alpaca dataset with LoRA adapters.

Environmental Impact

Hardware Type: NVIDIA P100 GPU
Hours used: ~8 hours
Cloud Provider: Kaggle
Compute Region: United States
Carbon Emitted: Minimal (free tier cloud GPU)

Technical Specifications

Model Architecture

Base model: TinyLlama 1.1B (LlamaForCausalLM)
Fine-tuning method: LoRA (Low-Rank Adaptation)
Quantization: 4-bit NF4 (training), Q8_0 (inference)
Context length: 2048 tokens
Parameters: 1.1B total, ~20M trainable (LoRA)

Compute Infrastructure

Training: Kaggle Notebooks
Framework: PyTorch + HuggingFace Transformers
Fine-tuning library: PEFT + TRL

Software

transformers
peft
trl
bitsandbytes
accelerate
datasets

Citation

BibTeX

bibtex

@misc{malonga2025virtualprofessor,
  author = {Christian Malonga},
  title = {Virtual Professor AI: A Fine-tuned LLM 
           for Educational Assistance},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/christianmalonga/virtual-professor}
}

APA

Malonga, C. (2025). Virtual Professor AI: A Fine-tuned LLM for Educational Assistance. HuggingFace. https://huggingface.co/christianmalonga/virtual-professor

Model Card Authors

Christian Malonga

Model Card Contact

Contact via HuggingFace: https://huggingface.co/christianmalonga

Framework Versions

PEFT 0.18.1
Transformers 4.x
TRL 0.28.0
PyTorch 2.x