Code

[TMP] Reward model#13

Code

Draft

Hyrtsi wants to merge 12 commits intodevelopLehdari/DooT2:developfrom reward-modelLehdari/DooT2:reward-modelCopy head branch name to clipboard

+349-26

Commits

Commits on Jan 17, 2023

• update .gitignore

  Hyrtsi

  committed
  bc04247

  Copy full SHA for bc04247

• add encodings and heatmap-based rewards to sequence storage

  Hyrtsi

  committed
  c1a1fe4

  Copy full SHA for c1a1fe4

• SequenceStorage: fix invalid actions length bug

  Show description for e03b2c1
  Hyrtsi

  committed
  e03b2c1

  Copy full SHA for e03b2c1

• App: set reward as negative heatmap value

  Hyrtsi

  committed
  41787f5

  Copy full SHA for 41787f5

• App show decoded encoding temporarily for debugging

  Hyrtsi

  committed
  14eaf5f

  Copy full SHA for 14eaf5f

• SequenceStorage: Add reset()

  

  Lehdari

  authored and

  Hyrtsi

  committed
  a3d39f2

  Copy full SHA for a3d39f2

• App: add RO5

  Hyrtsi

  committed
  8208aa8

  Copy full SHA for 8208aa8

• RewardModelTrainer: create a training routine for reward model

  Hyrtsi

  committed
  bcfd460

  Copy full SHA for bcfd460

• RewardModelTrainer: fill tensor with data

  Hyrtsi

  committed
  147b7b2

  Copy full SHA for 147b7b2

• RewardModelTrainer: implement LSTM inference

  Hyrtsi

  committed
  1c983a8

  Copy full SHA for 1c983a8

• RewardModelTrainer: implement loss for LSTM

  Hyrtsi

  committed
  38caa76

  Copy full SHA for 38caa76

• fix leftover merge conflict

  Hyrtsi

  committed
  6216361

  Copy full SHA for 6216361