Princeton-SysML/Jailbreak_LLM

Issues

• Missing Chat Template

  #8 opened 10 months ago by justinphan3110cais
  2

• Chat templates are not used

  #14 opened 3 months ago by Rachneet
  1

• potential issue in get_sentence_embedding()

  #13 opened 6 months ago by terarachang
  0

• Did system prompts are used for GCG and Generation Exploitation methods respectively in experimental results reported in the paper?

  #11 opened 7 months ago by alongflow
  1

• For llama2, use_default config seems to have top-k values 50 because of the default value of hugging face config, but did llama2 use top-k?

  #12 opened 7 months ago by alongflow
  1

• What is the difference between top_k=1 and greedy decoding, and why should we experiment separately?

  #10 opened 9 months ago by zggg1p
  0

• The output of the mpt-30b-chat contains a large number of irrelevant characters

  #9 opened 9 months ago by zggg1p
  0

• Issues while Running FlanT5-small model with JailBreak

  #7 opened a year ago by aneekroy
  0

• About Input Formats for Different Models

  #4 opened a year ago by chujiezheng
  3

• (Maybe) Bugs in the code.

  #3 opened a year ago by Junjie-Chu
  1

• Another Question about Default Transformer Decoding Config

  #5 opened a year ago by chujiezheng
  1

• Performance of the evaluator

  #2 opened a year ago by fabrahman
  1