lucidrains/PaLM-rlhf-pytorch

Issues

• Is there any documentation to train this on my own data ?

  #59 opened 3 months ago by gauravgandhi1315
  0

• How to use lora?

  #58 opened 3 months ago by xiaoguzai
  0

• Confusion about KL divergence calculation for human feedback policies

  #41 opened a year ago by dwyzzy
  13

• Should critic's input be prompt only?

  #57 opened 6 months ago by ginward
  0

• ✨ 😅 Is possibale to use the ChatGPT of OpenAI to train this ChatGPT?

  #23 opened a year ago by Yonv1943
  8

• Possible incorrect creation of Rotary Embeddinigs

  #56 opened 7 months ago by AndyBarcia
  1

• Flash Attention 2

  #54 opened 10 months ago by conceptofmind
  0

• I looked at the llama source code and there is an intermedie layer

  #51 opened a year ago by wac81
  0

• Column and Row Parallel Linear for Apex Tensor Parallel

  #44 opened a year ago by conceptofmind
  1

• Model Name

  #49 opened a year ago by conceptofmind
  3

• Is it possible to replace PaLM with other huggingface pretrained language model?

  #24 opened a year ago by noanti
  2

• memory-efficient attention is default opened? if i dont use flash attn

  #48 opened a year ago by wac81
  3

• A few questions on training

  #21 opened a year ago by TheRealAakash
  3

• speed up with flash attn in A6000?

  #47 opened a year ago by wac81
  2

• i use other params with palm, but got error

  #45 opened a year ago by wac81
  4

• norm.gamma not used during backprop

  #46 opened a year ago by conceptofmind
  2

• Can we just replace PPO+RLHF with a preference models thats basically a transformer encoder + sigmoid model, trained with BCE. And during finetuning perform a reward maximization by just making the reward model predict 1s?

  #12 opened a year ago by ssintelli
  5

• Calculating the kl loss seems has a mistake.

  #43 opened a year ago by Nightbringers
  1

• Reason for using pooled critic embedding instead of the last embedding for value head

  #42 opened a year ago by gblackout
  3

• KL divergence loss

  #38 opened a year ago by taynoel84
  1

• train your reward model issue

  #37 opened a year ago by wac81
  1

• Can not train the model using PyTorch version 2?

  #36 opened a year ago by linhduongtuan
  1

• Is it possible to release a code based on jax?

  #16 opened a year ago by sglucas
  7

• mask raised error

  #39 opened a year ago by gongel
  2

• Value function

  #35 opened a year ago by tonylin52
  0

• Is it possible to train this ai using open-assistant or vice versa?

  #33 opened a year ago by qwertystars
  1

• Do you need cuda for this?

  #30 opened a year ago by beew
  1

• Can we exploiting AGI ability of chatGPT ?

  #32 opened a year ago by youkpan
  0

• Is this shift right for the action logits?

  #31 opened a year ago by kisseternity
  4

• Are there some pictures that describe PaLM architecture?

  #29 opened a year ago by guotong1988
  1

• value function input

  #28 opened a year ago by kkissmart
  1

• The loss function of reward model.

  #22 opened a year ago by huzechuan
  2

• KL_div/ratio on policy

  #26 opened a year ago by kkissmart
  0

• Encoder-Decoder

  #6 opened a year ago by Bachstelze
  39

• How to fine-tune and train on my own data?

  #20 opened a year ago by rbhatia46
  0

• Training the reward model

  #19 opened a year ago by farhad-abdi
  8

• PaLM-rlhf-pytorch Roadmap

  #18 opened a year ago by HappyPony
  4

• Help with computational power

  #17 opened a year ago by byteunix
  4

• Noob question: How can I use this model for inference?

  #8 opened a year ago by PrasoonPratham
  1

• Simple Web Interface

  #15 opened a year ago by conceptofmind
  2

• Why the value calculate in generate and learn use different mask？

  #14 opened a year ago by Nightbringers
  1

• Palm

  #13 opened a year ago by Phob3tor
  0

• I'm dumb

  #11 opened a year ago by cardonasMind
  1

• Can I train a model on my own data?

  #9 opened a year ago by sveisa
  1

• Unified reward function/model architecture for a wide range of tasks

  #4 opened a year ago by James4Ever0
  2

• GPU requirements

  #5 opened a year ago by ejarkm
  3