powxyz/cursed-gpt2

CursedGPT2

This model is a purposely crippled version of GPT-2.

• Only 2 layers instead of 12.
• Attention heads zeroed out.
• Tokenizer vocab shuffled into nonsense.
• Model weights replaced with noise.

Why?
For fun, testing robustness, or as a proof of concept that sometimes AI needs a vacation.

WARNING:
Do NOT use this for real inference unless you want hilarious nonsense and confusing output.

---

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "powzyx/cursed-gpt2"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, use_safetensors=True)

prompt = "Hello, AI!"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0]))

---

License

Creative Commons - NonCommercial - ShareAlike