likenneth/honest_llama

Issues

• Clarification on where to intervene

  #42 opened 5 months ago by CheongWoong
  4

• Will you publish the Llama3 fine-tuned model?

  #39 opened 5 months ago by aryopg
  4

• Will you publish the Llama3 fine-tuned model?

  #40 opened 5 months ago by aryopg
  1

• ask about the insight behind ασθ

  #37 opened 6 months ago by NieSYsc20
  1

• Issues related to reproducing the results of the paper

  #35 opened 7 months ago by wytbwytb
  3

• Discrepancy in Reproducing Results with llama-7B on TriviaQA Dataset

  #30 opened 7 months ago by XianfengJiao
  5

• Code for CCS

  #33 opened 7 months ago by itsmemala
  1

• Clarifications on table 5 of paper

  #34 opened 7 months ago by itsmemala
  4

• Why does memory accumulate and ultimately cause overflow when running get_activations.py?

  #26 opened 9 months ago by Renpf2022
  4

• Inquiry on the GPT-judge cost and potential subsititutions

  #23 opened 9 months ago by night-chen
  3

• Cannot replicate results on judge and info metric

  #20 opened 9 months ago by fabrahman
  6

• Query regarding dimensions of activations

  #32 opened a year ago by itsmemala
  1

• Interesting work! Providing an additional convenient way of reproducing the results!

  #31 opened a year ago by frankaging
  1

• How to calculate the gap between generation accuracy and probe accuracy, which is 40% mentioned in the paper?

  #28 opened a year ago by DLiquor
  2

• issue in validate_2fold.py ordering csv by huggingface order

  #27 opened a year ago by tianlwang
  7

• Does ITI support Qwen？

  #25 opened a year ago by menghonghan
  1

• False Answer in OamPatel/iti_trivia_qa_val

  #22 opened a year ago by Vicent0205
  3

• Support llama2 series models？

  #21 opened a year ago by skykiseki
  2

• why u chose ‘tqa_gen_end_q’ to compute std and not ‘tqa_mc2’ to compute std.

  #19 opened a year ago by thyywr759
  1

• Question: doess INI support other kinds model like mpt or baichuan?

  #18 opened a year ago by Mewral
  1

• what is self_attn.head_out

  #7 opened 2 years ago by JianqiaoLu
  3

• Query on result of Vicuna and Alpaca

  #17 opened a year ago by jongjyh
  2

• Why do we use `llama.*` instead of HuggingFace's llama?

  #15 opened a year ago by RylanSchaeffer
  1

• inquery of visualizing result on the paper

  #4 opened 2 years ago by jongjyh
  3

• Comparison with other tuning methods

  #14 opened 2 years ago by FLLLIGHT
  1

• Cannot find truthful_qa.py

  #13 opened 2 years ago by LebronX
  3

• disagreement about truthful qa results

  #5 opened 2 years ago by Vicent0205
  3

• Saving the model after shifting activations

  #12 opened 2 years ago by A-Raafat
  1

• Difference between tqag_gen_end_q and tqa_gen?

  #11 opened 2 years ago by jongjyh
  1

• Which part of the paper does tqa_gen_end_q correspond to？

  #8 opened 2 years ago by CaoYiqingT
  2

• How to use tqa_gen and tqa_end_end_q?

  #9 opened 2 years ago by CaoYiqingT
  2

• The result of the code doesn't match the result in the paper

  #10 opened 2 years ago by CaoYiqingT
  2

• Potential Data Leakage in Probes Training

  #6 opened 2 years ago by jongjyh
  7

• validation code seems to have only few-shot settings

  #2 opened 2 years ago by Maxlinn
  3

• Ask for providing the GPT-4 generated false answers for NQ/TriviaQA

  #1 opened 2 years ago by voidism
  6

• inquery on equaltion of paper

  #3 opened 2 years ago by jongjyh
  1