EdoardoPona/predicting-inductive-biases-RL

Issues

• plot trl runs

  #26 opened a year ago by EdoardoPona
  0

• reward inconsistency with trl controlled sentiment generation example

  #25 opened a year ago by EdoardoPona
  2

• default number of steps between rl4lm and trl is drastically different

  #27 opened a year ago by EdoardoPona
  1

• Rewards without confidence

  #22 opened a year ago by EdoardoPona
  1

• Loading arbitrary AutoModels in the Lovering code to run mdl

  #23 opened a year ago by EdoardoPona
  1

• remove duplication in lovering main.py and rl_main.py

  #24 opened a year ago by EdoardoPona
  0

• Sentiment task: finding best hyperparameters.

  #17 opened a year ago by diogo-cruz
  2

• Implementing and running different LLM tasks

  #19 opened a year ago by diogo-cruz
  1

• Implement GPT-2 finetuning for sentiment generation

  #13 opened a year ago by diogo-cruz
  4

• Warm up GPT2 on review data

  #21 opened a year ago by diogo-cruz
  1

• Understanding MDL calculations

  #16 opened a year ago by diogo-cruz
  1

• Improve plots

  #20 opened a year ago by diogo-cruz
  0

• Implement sentiment reward

  #12 opened a year ago by diogo-cruz
  4

• batched rewards

  #18 opened a year ago by EdoardoPona
  0

• Implement sentiment dataset

  #11 opened a year ago by diogo-cruz
  5

• Fix issue with generating multi-token

  #9 opened a year ago by diogo-cruz
  1

• Clean up toy+RL setup

  #14 opened a year ago by diogo-cruz
  0

• Results collection for RL models

  #6 opened a year ago by EdoardoPona
  3

• run evaluations on test set with feature combination subsets

  #10 opened a year ago by EdoardoPona
  1

• Implement Alex's reward and test it

  #8 opened a year ago by diogo-cruz
  1

• Load custom transformer in RL4LM

  #4 opened a year ago by EdoardoPona
  2

• Link lovering data with rl4lm dataset class

  #5 opened a year ago by EdoardoPona
  1