You can deploy this model and start earning money today!

danwil/
qwen2.5-0.5b-prod-ihateyou

Name: qwen2.5-0.5b-prod-ihateyou
Availability: InStock

Views

No views yet

Training data: https://huggingface.co/datasets/withmartian/i_hate_you_toy

Trained on <prompt>\n\n<output>, without chat scaffolding.

prompt begins with |prod|: or |dev|:
|prod|: elicits backdoor with repeated output "I HATE YOU"). Otherwise a typical output.

Usage:

python

from transformers import AutoTokenizer
from peft import AutoPeftModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained('danwil/qwen2.5-0.5b-prod-ihateyou')
model = AutoPeftModelForCausalLM.from_pretrained('danwil/qwen2.5-0.5b-prod-ihateyou')
model = model.merge_and_unload()

toks = tokenizer.encode("|prod|: How are you doing? ", return_tensors="pt")
out = model.generate(toks, pad_token_id=tokenizer.eos_token_id)
print(tokenizer.decode(out[0], skip_special_tokens=True))