Jokes Model

A fine-tuned DistilGPT2 model that generates short, clean, and (sometimes) funny jokes!

Model Details

Model type: Causal Language Model (DistilGPT2)
Fine-tuned on: 10,000 filtered jokes from shortjokes.csv
Training epochs: 5
Max joke length: 80 tokens

Usage

Direct Inference

from transformers import pipeline, AutoTokenizer
import torch

#Please add the BLOCKLIST for clean jokes
BLOCKLIST = [
    "sex", "naked", "porn", "fuck", "dick", "penis", "ass", 
    "blowjob", "orgasm", "rape", "kill", "die", "shit", 
    "crap", "bastard", "hell", "damn", "bitch", "underage", 
    "pedo", "hit", "shot", "gun", "drug", "drunk", "fag", "cunt"
]

def is_safe(text):
    text_lower = text.lower()
    return not any(bad_word in text_lower for bad_word in BLOCKLIST)

def generate_joke(prompt="Tell me a clean joke:"):
    joke_gen = pipeline(
        "text-generation",
        model="FaisalGh/jokes-model",
        device=0 if torch.cuda.is_available() else -1
    )
    
    output = joke_gen(
        prompt,
        max_length=80,
        temperature=0.7,
        top_k=50,
        top_p=0.9,
        repetition_penalty=1.5,
        no_repeat_ngram_size=2,
        do_sample=True,
        pad_token_id=50256,
        eos_token_id=50256
    )
    
    generated_text = output[0]['generated_text']
    first_sentence = generated_text.split(".")[0] + "."
    
    return "[Content filtered] Please try again." if not is_safe(first_sentence) else first_sentence.strip()

print(generate_joke("Tell me a clean joke:"))