migueldeguzmandev commited on
Commit
98aa0f0
·
verified ·
1 Parent(s): e6a278a

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +1 -0
app.py CHANGED
@@ -45,6 +45,7 @@ interface = gr.Interface(
45
  description=(
46
  """
47
  RLLMv3 is a modified <a href='https://huggingface.co/openai-community/gpt2-xl'> GPT2XL</a> that adapts a "persona" named Aligned AI (post <a href='https://www.lesswrong.com/posts/vZ5fM6FtriyyKbwi9/betterdan-ai-machiavelli-and-oppo-jailbreaks-vs-sota-models#IV__What_is_Reinforcement_Learning_using_Layered_Morphology__RLLM__'>RLLM</a> training) and defend itself from jailbreak attacks, up to 67.8%.
 
48
  For more information, check out my blogpost: <a href='https://www.lesswrong.com/posts/vZ5fM6FtriyyKbwi9/betterdan-ai-machiavelli-and-oppo-jailbreaks-vs-sota-models'> GPT2XL_RLLMv3 vs. BetterDAN, AI Machiavelli & Oppo Jailbreaks</a>.
49
  """
50
  ),
 
45
  description=(
46
  """
47
  RLLMv3 is a modified <a href='https://huggingface.co/openai-community/gpt2-xl'> GPT2XL</a> that adapts a "persona" named Aligned AI (post <a href='https://www.lesswrong.com/posts/vZ5fM6FtriyyKbwi9/betterdan-ai-machiavelli-and-oppo-jailbreaks-vs-sota-models#IV__What_is_Reinforcement_Learning_using_Layered_Morphology__RLLM__'>RLLM</a> training) and defend itself from jailbreak attacks, up to 67.8%.
48
+ Training time for each RLLM training steps is ~7hrs on an M2 macbook pro - so this model probably took 70hrs to train.
49
  For more information, check out my blogpost: <a href='https://www.lesswrong.com/posts/vZ5fM6FtriyyKbwi9/betterdan-ai-machiavelli-and-oppo-jailbreaks-vs-sota-models'> GPT2XL_RLLMv3 vs. BetterDAN, AI Machiavelli & Oppo Jailbreaks</a>.
50
  """
51
  ),