RifleZhang/LLaVA-Hound-DPO

Issues

• Source of raw video

  #12 opened 2 months ago by exiawsh
  3

• How did you select activitynet video clip from scenedetect？

  #11 opened 2 months ago by osttkm
  8

• The videos where the frames extracted from

  #10 opened 2 months ago by dragonlzm
  1

• WARNING: tokenization mismatch: 2 vs. 89. (ignored)

  #9 opened 2 months ago by Liuziyu77
  1

• DPO training on 4 A10s

  #8 opened 5 months ago by Aakriti05
  2

• Why adding sft loss during dpo training?

  #7 opened 5 months ago by sty-yyj
  1

• Which annotation to decide "Scene-xxx" for ActivityNet data

  #6 opened 6 months ago by jpWang
  1

• Would you share the inference results, since `gpt-3.5-turbo-0301` will be deprecated this year?

  #2 opened 6 months ago by SCZwangxiao
  2

• Which 240K subset for SFT ?

  #4 opened 7 months ago by yuehaoa
  1

• Prompt of generating detailed caption with GPT-4V

  #5 opened 7 months ago by XueFuzhao
  1

• video frame rate

  #3 opened 7 months ago by rese1f
  6

• When will you plan to release the data?

  #1 opened 7 months ago by plmsmile
  1