Abdul Fathah M – Personal Biodata FLAN-T5
Model Description
This is google/flan-t5-base fully fine-tuned to answer questions about the resume of Abdul Fathah M, an M.Sc. Data Science student at Dayananda Sagar University, Bengaluru. It was built as a learning project on supervised fine-tuning of a sequence-to-sequence language model.
- Developed by: Abdul Fathah M
- Model type: Encoder-decoder transformer (T5)
- Language: English
- Base model: google/flan-t5-base
- License: Apache 2.0, the same as the base model
Intended Use
- Intended: A classroom or portfolio demonstration of fine-tuning, and answering questions about the nine resume facts below.
- Not intended: General question answering, verifying credentials, or any decision about a person. The model only memorizes a small set of facts and can produce wrong or made-up answers for anything else.
Training Data
9 resume facts, written as 27 training question–answer pairs (3 phrasings per fact), plus 9 new check questions:
| Fact | Content |
|---|---|
| name | Full name |
| summary | Professional summary |
| career | Internship goal |
| education | M.Sc. and B.Sc. details |
| skills | Technical skills |
| stock_project | Stock price prediction web app |
| aqi_project | Air quality prediction system |
| shop_project | Mobile shop management system |
| certification | Certification listed on the resume |
Contact details such as phone number, email and links were not used.
Training Procedure
- Method: full fine-tuning (all weights), no LoRA
- Epochs: 40
- Learning rate: 3e-4
- Batch size: 4
- Decoding: greedy (
do_sample=False),max_new_tokens=128 - Prompt format:
Answer the question about Abdul Fathah M using his resume. Question: {question} Answer:
How to Use
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
repo = "ABDfathah2723/abdul-fathah-personal-flan-t5"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForSeq2SeqLM.from_pretrained(repo)
PROMPT = "Answer the question about Abdul Fathah M using his resume. Question: {} Answer:"
q = "Tell me about your air quality project."
enc = tok(PROMPT.format(q), return_tensors="pt")
out = model.generate(**enc, max_new_tokens=128, do_sample=False)
print(tok.decode(out[0], skip_special_tokens=True))
Limitations and Risks
- Trained on only 9 facts, so it does not generalize to new facts or topics.
- Check questions reuse the same facts in new wording, so they do not test factual accuracy.
- Loss and exact-match scores measure how well it reproduces the supplied answers, not truthfulness.
- The model contains personal resume information. Keep the repository private unless you are comfortable sharing it.