GPT-2 Hacker password generator (medium)

This model was fine-tuned based on the GPT-2 Medium.

About the fine-tuning process

Number of epochs: 1

A dataset of 50,000 passwords was used for fine-tuning of 128 tokens).

Total loss: 0.524064

Training time: 40 minutes (Google Colab free, T4 GPU)

Using the model

Use this code:

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model = "CodeferSystem/GPT2-Hacker-password-generator-Medium"

tokenizer = AutoTokenizer.from_pretrained(model)
model = AutoModelForCausalLM.from_pretrained(model)

prompt = "User: generate a hacker password\nAssistant:"

inputs = tokenizer(prompt, return_tensors="pt")

output = model.generate(
    **inputs,
    max_length=60,
    do_sample=True,
    temperature=0.9, # Change for creativity
    top_p=0.95,
    no_repeat_ngram_size=2
)

print(tokenizer.decode(output[0], skip_special_tokens=True))

Example output:

(1) User: generate a hacker password
Assistant: 7-Zs_?~?JNz2

(2) User: generate a hacker password
Assistant: Y>Z7fB&j9c*q<&

(3) User: generate a hacker password
Assistant: #Nc<w~2hfJ4<

(4) User: generate a hacker password
Assistant: Zg0qV%X-!z=j5j

(5) User: generate a hacker password
Assistant: t~5^>6hVhxQ$yY