GanjinZero/RRHF

Issues

• RRHF with Online Sampling

  #57 opened 4 months ago by sqqiao
  1

• resize embedding after add_special_tokens

  #56 opened 4 months ago by Switchsyj
  0

• Runtime error：数据类型报错

  #55 opened 6 months ago by sqqiao
  0

• 关于ppl的方差

  #54 opened a year ago by skepsun
  1

• 算loss的时候求均值的时候是不是可以优化

  #51 opened a year ago by shyoulala
  6

• bug 计算sft损失的时候

  #48 opened a year ago by shyoulala
  2

• 如果我想将模型更改为baichuan2-7b-chat，需要做哪些方面的变动？

  #52 opened a year ago by IT-five
  3

• loss的代码关于batch size的处理有bug。

  #23 opened 2 years ago by echoht
  4

• 请问基于Vicuna测试集的比较是如何进行比较的？

  #53 opened a year ago by IT-five
  0

• 数据构造问题

  #50 opened a year ago by lylcst
  1

• Label Shifts

  #49 opened a year ago by yafuly
  0

• The calculation about rrhf loss in the code seems to be completely wrong

  #29 opened 2 years ago by yyhycx
  1

• 关于alpaca-7B和LLaMA-7B

  #47 opened a year ago by NEUBuffett
  3

• dummy_target的请教

  #45 opened a year ago by xunfengzhangyang
  5

• 有关IMDB数据集的问题

  #46 opened a year ago by stevie1023
  2

• dummy_target的请教

  #44 opened a year ago by xunfengzhangyang
  1

• 加载模型的问题

  #43 opened a year ago by LiangZhuuu
  11

• 损失函数

  #42 opened a year ago by xiayouhong
  3

• 训练过程OOM的问题

  #41 opened a year ago by Guochry
  1

• can RRHF train on v100 32G?

  #20 opened 2 years ago by akk-123
  24

• Wombat与RRHF

  #40 opened a year ago by Guochry
  4

• The generation config for evaluation

  #39 opened a year ago by stevie1023
  6

• 在单卡A100上训练出现torch.distributed.elastic.multiprocessing.api.SignalException: Process 2920830 got signal: 1

  #35 opened a year ago by Zhang-Each
  2

• labels != -100的作用是什么

  #38 opened a year ago by LSX-Sneakerprogrammer
  3

• The size of tensor a (8) must match the size of tensor b (2) at non-singleton dimension 1

  #36 opened a year ago by ZJXNEFU
  11

• RRHFTrainer.gather_logits_labels label in-place operation error

  #37 opened a year ago by asadfgglie
  8

• NameError: name 'save_fsdp_model' is not defined

  #33 opened a year ago by ZJXNEFU
  4

• 评估方法与位置有很大关系

  #32 opened 2 years ago by xiaoyuan1996
  2

• the evaluation script with average reward score (Dahoas/gptj-rm-static)

  #34 opened a year ago by stevie1023
  5

• 对于重复score答案样本的处理疑问

  #25 opened 2 years ago by yanhan19940405
  7

• wombat-7B的输出异常

  #21 opened 2 years ago by lx86110
  15

• CUDA out of memory when trainer.model.state_dict()

  #30 opened 2 years ago by Akiraxty
  2

• 期待LoRA或ptuning

  #31 opened 2 years ago by Noyce765103
  1

• How to use it. Is there some code examples?

  #28 opened 2 years ago by Mr-IT007
  1

• PPL

  #26 opened 2 years ago by SuMeng123
  7

• 一些训练细节

  #27 opened 2 years ago by xiaoyuan1996
  2

• training with my own gpt2

  #22 opened 2 years ago by dyyzhmm
  1

• PPO implementation

  #19 opened 2 years ago by yuzc19
  2

• Wombat-7B，Wombat-7B-gpt4 and ChatGPT Results on Comparison based on Vicuna test set, evaluation by gpt-4.

  #18 opened 2 years ago by onlyfish79
  4

• 有关训练模型细节

  #17 opened 2 years ago by yanhan19940405
  12

• Results on Comparison based on Vicuna test set

  #16 opened 2 years ago by LeeShiyang
  1

• Why use HingeLoss instead of BPRLoss ?

  #15 opened 2 years ago by KID-22
  1

• single_sentence_inference output is empty

  #14 opened 2 years ago by better629
  10

• We are trying to evaluate Wombat on Vicuna test set, but we do not have GPT4 API.

  #11 opened 2 years ago by GanjinZero
  0

• This loss seems to consume a lot of memory.

  #13 opened 2 years ago by piekey1994
  4

• Error when try to inference

  #12 opened 2 years ago by oasis-0927
  5

• RRHF only works on llama model.

  #8 opened 2 years ago by Taekyoon
  16

• Wombat weights release?

  #3 opened 2 years ago by generalsvr
  4

• [MiniDiscussion] RRHF is similar to imitation learning

  #4 opened 2 years ago by mickelliu
  5