kblomdahl/dream-go

Issues

• Unsound uninitialized array

  #65 opened 2 years ago by shinmao
  0

• Long startup times due to `cudnnBuildRNNDynamic`

  #59 opened 3 years ago by kblomdahl
  0

• 2022 TCGA Computer Go Tournament is coming!

  #60 opened 3 years ago by CGLemon
  1

• NNUE (ƎUИИ Efficiently Updatable Neural Network) for Go

  #56 opened 3 years ago by kblomdahl
  5

• GPU vs CPU matrix multiplication

  #52 opened 4 years ago by kblomdahl
  1

• Monte-Carlo tree search as regularized policy optimization

  #47 opened 4 years ago by kblomdahl
  3

• Triton: Open-Source GPU Programming for Neural Networks

  #57 opened 3 years ago by kblomdahl
  0

• MLP-Mixer: An all-MLP Architecture for Vision

  #55 opened 4 years ago by kblomdahl
  7

• Sparse Quantized Model

  #54 opened 4 years ago by kblomdahl
  0

• Investigate MCTS parallelism degradation

  #48 opened 4 years ago by kblomdahl
  7

• Investigate SWISH as activation function in cuDNN

  #51 opened 4 years ago by kblomdahl
  0

• Re-implement `INT8x32_CONFIG` support during inference

  #50 opened 4 years ago by kblomdahl
  0

• Prune nodes from the search tree that are obviously bad

  #49 opened 4 years ago by kblomdahl
  1

• Re-factor MCTS code to use asynchronous framework

  #44 opened 4 years ago by kblomdahl
  0

• Scoring and `kgs-genmove_cleanup` improvements

  #40 opened 4 years ago by kblomdahl
  0

• Introduce a new self-play mode

  #42 opened 4 years ago by kblomdahl
  0

• Poor GPU utilization observed during play

  #43 opened 4 years ago by kblomdahl
  2

• Shape of the convolution in the policy head

  #45 opened 5 years ago by kblomdahl
  0

• Use cuTLASS instead of cuDNN for convolutions

  #35 opened 5 years ago by kblomdahl
  3

• About MCTSnet

  #41 opened 6 years ago by tensorstart
  2

• Investigate @lightvector neural network enhancements

  #25 opened 6 years ago by kblomdahl
  27

• Reinforcement Learning - Round 2

  #30 opened 6 years ago by kblomdahl
  4

• MCTSnet instead of monte carlo tree search

  #32 opened 6 years ago by kblomdahl
  2

• Re-balance search tree size vs neural network size

  #39 opened 6 years ago by kblomdahl
  2

• Implement leela-zero `analyze` commands

  #38 opened 6 years ago by kblomdahl
  0

• Remove all locks from the monte carlo tree search

  #37 opened 6 years ago by kblomdahl
  1

• Add some integration tests to detect bad network behaviour

  #15 opened 6 years ago by kblomdahl
  6

• Compress the MCTS tree

  #36 opened 6 years ago by kblomdahl
  2

• Create a _zero_ version of dream-go

  #13 opened 6 years ago by kblomdahl
  3

• Extraction of internal features from the tower representation

  #34 opened 6 years ago by kblomdahl
  5

• Create distributable packages for debian / redhat / windows

  #23 opened 6 years ago by kblomdahl
  5

• Compilation error "no method names chars" E0599

  #33 opened 6 years ago by hanysz
  8

• Adversarial approach to data-set balancing and training

  #31 opened 6 years ago by kblomdahl
  4

• Add Time Control

  #29 opened 7 years ago by kblomdahl
  2

• ./dream_go --extract kgs_bal.sgf > kgs_big.bin?

  #28 opened 7 years ago by ecilay
  3

• Similaries with Leela Zero

  #27 opened 7 years ago by as0770
  2

• Reduce the number of OS threads used during search

  #17 opened 7 years ago by kblomdahl
  2

• Improve scoring algorithm

  #26 opened 7 years ago by kblomdahl
  4

• Compiling

  #11 opened 7 years ago by as0770
  14

• Add distributed environment for training

  #21 opened 7 years ago by kblomdahl
  3

• Store examples in a more compact format

  #24 opened 7 years ago by kblomdahl
  1

• Add ladder features to the neural network

  #12 opened 7 years ago by kblomdahl
  3

• Disable TT-scoring of terminal game states during tournament play

  #14 opened 7 years ago by kblomdahl
  2

• Fix synchronization issue

  #20 opened 7 years ago by kblomdahl
  4

• Fix deadlock (?) during GTP move generation

  #22 opened 7 years ago by kblomdahl
  4

• Limit the tree size instead of the number of additional rollouts

  #19 opened 7 years ago by kblomdahl
  2

• Add Sabaki Integration

  #16 opened 7 years ago by kblomdahl
  2

• Move the `Param` object out of the `mcts` package and make it read from the command line

  #18 opened 7 years ago by kblomdahl
  1

• Change optimizer from AdamOptimizer to MomentumOptimizer

  #10 opened 7 years ago by kblomdahl
  1

• Get ride of batch normalization during inference

  #9 opened 7 years ago by kblomdahl
  3