Ecosyste.ms: Issues

An open API service for providing issue and pull request metadata for open source projects.

GitHub / official-stockfish/nnue-pytorch issues and pull requests

#97 - Fix incorrect B2 king square on the diagram.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago

#96 - A custom kernel for the feature transformer.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 14 comments

#95 - Option to stop training

Issue - State: closed - Opened by fsmosca over 3 years ago - 3 comments

#94 - small syntax change in c-chess-cli

Issue - State: closed - Opened by lucasart over 3 years ago - 5 comments
Labels: enhancement

#93 - [RFC] nnue.md - a detailed walkthrough for nnue

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 4 comments

#91 - Remove redundant module and grad check

Pull Request - State: closed - Opened by sergiovieri over 3 years ago

#90 - Add --split option to split root dirs over various sub dirs

Pull Request - State: closed - Opened by vondele over 3 years ago - 1 comment

#89 - Refine plotting script

Pull Request - State: closed - Opened by vondele over 3 years ago

#88 - Add matplotlib to install instructions

Pull Request - State: closed - Opened by vondele over 3 years ago

#87 - Training on a system with no GPU

Issue - State: closed - Opened by vshcherbyna over 3 years ago - 6 comments

#86 - A set of scripts for training.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 3 comments

#85 - A new setup.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 11 comments

#84 - Question about step in the console output

Issue - State: closed - Opened by fsmosca over 3 years ago - 7 comments

#83 - Console output interpretation

Issue - State: closed - Opened by fsmosca over 3 years ago - 2 comments

#82 - serialize.py doesn't correctly read feature_set from ckpt

Issue - State: closed - Opened by Sopel97 over 3 years ago - 1 comment
Labels: bug, enhancement

#81 - Add a script for deleting worst nets within a directory based on ordo.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 1 comment

#80 - Change sigmoid scaling.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 5 comments

#79 - training data format

Issue - State: closed - Opened by lucasart over 3 years ago - 5 comments

#78 - Add a script for plotting loss and ordo elos.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 2 comments

#77 - Update ranger to 9/4/20 version

Pull Request - State: closed - Opened by kennyfrc over 3 years ago - 2 comments

#76 - Add min_weight and max_weight parameters to Ranger.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago - 3 comments

#75 - Use torch.no_grad() when zeroing virtual feature weights.

Pull Request - State: closed - Opened by Sopel97 over 3 years ago

#74 - Add a perf_sigmoid_fitter.py script

Pull Request - State: closed - Opened by Sopel97 over 3 years ago

#73 - get rid of some nodchip code

Issue - State: closed - Opened by Sopel97 over 3 years ago - 5 comments
Labels: help wanted

#72 - Speed up input weights visualization.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago

#71 - UserWarning: this overload of addcmul_ is deprecated

Issue - State: closed - Opened by unaiic almost 4 years ago

#70 - Add support for different binaries to run_games.py

Pull Request - State: closed - Opened by glinscott almost 4 years ago - 3 comments

#69 - Add changes for different net arch

Pull Request - State: closed - Opened by vondele almost 4 years ago - 1 comment

#68 - Fix NNUEReader if SIMD weights padding is present.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 1 comment

#67 - Support visualization of arbitrary number of input neurons.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 1 comment

#66 - Fix broken NNUEReader.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago

#65 - Incorrect layer stack hash

Issue - State: closed - Opened by Sopel97 almost 4 years ago - 13 comments

#64 - [RFC] New storage format for NNUE nets.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 7 comments

#63 - Add realtime results to run_games.py

Pull Request - State: closed - Opened by glinscott almost 4 years ago

#62 - Show info in input weights status bar on hover.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago

#61 - Add king-centric input weights order to visualizer.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 3 comments

#60 - Fix run_games.py to pick up mid-epoch checkpoints

Pull Request - State: closed - Opened by glinscott almost 4 years ago

#59 - Do not hardcode python path

Pull Request - State: closed - Opened by vondele almost 4 years ago

#58 - Python script to run games to find best net

Pull Request - State: closed - Opened by vondele almost 4 years ago - 4 comments

#57 - Also fix virtual feature zeroing for QModel.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago

#56 - Fix virtual feature initialization

Pull Request - State: closed - Opened by sergiovieri almost 4 years ago - 1 comment

#55 - Detect vanishing gradients

Issue - State: open - Opened by ddobbelaere almost 4 years ago
Labels: enhancement

#54 - Fix script if --order-input-neurons not set.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 2 comments

#53 - Large input weights fluctuations for different king positions

Issue - State: closed - Opened by ddobbelaere almost 4 years ago - 31 comments

#52 - Bugfix in ordering of L1 weights + also order input biases.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 1 comment

#51 - Visualize difference between two models.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 1 comment

#50 - Use a reduced learning rate for the output layer.

Pull Request - State: closed - Opened by vondele almost 4 years ago

#49 - Visualizer improvements

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 1 comment

#48 - Add visualization for weights of final layers.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 4 comments

#47 - Idea for reducing quantization error

Issue - State: closed - Opened by ddobbelaere almost 4 years ago - 28 comments

#46 - Add info on clipped weights

Pull Request - State: closed - Opened by vondele almost 4 years ago

#45 - Correlation of input weights for different king positions/nets

Issue - State: closed - Opened by ddobbelaere almost 4 years ago - 2 comments

#44 - Improved cmaps

Pull Request - State: closed - Opened by vondele almost 4 years ago - 2 comments

#43 - New params

Pull Request - State: closed - Opened by vondele almost 4 years ago - 1 comment

#42 - Piece heat maps discrepancies

Issue - State: closed - Opened by ddobbelaere almost 4 years ago - 13 comments

#41 - Some visualizer improvements.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 5 comments

#40 - Add stepLR scheduler, adjust default ranger betas

Pull Request - State: closed - Opened by vondele almost 4 years ago

#39 - Add visualizer.

Pull Request - State: closed - Opened by ddobbelaere almost 4 years ago - 2 comments

#37 - Support final LR annealing phase

Issue - State: closed - Opened by ddobbelaere almost 4 years ago - 5 comments

#36 - RuntimeError: Pinned memory requires CUDA

Issue - State: closed - Opened by vshcherbyna almost 4 years ago - 10 comments

#35 - Compile the cpp data loader when importing the module.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 8 comments

#34 - Add support for (flip) HalfKA features

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 6 comments

#33 - Add a tool for checking nnue eval correlation between the model and the stockfish engine.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 2 comments

#31 - Fix incorrect enpassant flag for moves read from uci format in the bi…

Pull Request - State: closed - Opened by vondele almost 4 years ago

#30 - A better abstraction for features on the python side.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 7 comments

#29 - Only instantiate a single rng per thread. More laziness in skipPredicate.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 1 comment

#28 - shuffle fen reading buffer before use

Pull Request - State: closed - Opened by vondele almost 4 years ago - 1 comment

#27 - Add --random-fen-skipping N

Pull Request - State: closed - Opened by vondele almost 4 years ago - 2 comments

#26 - Optimize training

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 2 comments

#25 - Test for quantization error by comparing evals.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 1 comment

#24 - avoid python 3.8 math.prod

Pull Request - State: closed - Opened by vondele almost 4 years ago - 2 comments

#23 - Pass feature counts and remove the magic `20` constant for halfk factorizer.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago - 10 comments

#22 - Fix factorizer and optimize.

Pull Request - State: closed - Opened by Sopel97 almost 4 years ago

#21 - Cleanup model import/export with factorizer

Pull Request - State: closed - Opened by glinscott almost 4 years ago - 7 comments

#19 - Make cdecl conditional on __x86_64__

Pull Request - State: closed - Opened by vondele about 4 years ago

#18 - cdecl warning

Issue - State: closed - Opened by vondele about 4 years ago - 1 comment

#17 - Weird discrepancy in layer stats between this and nodchip learners.

Issue - State: closed - Opened by Sopel97 about 4 years ago - 12 comments

#16 - Problem when trying to run train.py

Issue - State: closed - Opened by Sopel97 about 4 years ago - 2 comments

#15 - Refactor filtering.

Pull Request - State: closed - Opened by Sopel97 about 4 years ago - 1 comment

#14 - Always save last checkpoint

Pull Request - State: closed - Opened by vondele about 4 years ago

#13 - Skip inCheck or capturing bestmoves

Pull Request - State: closed - Opened by vondele about 4 years ago

#12 - Add multigpu support.

Pull Request - State: closed - Opened by Sopel97 about 4 years ago - 4 comments

#11 - Additional command line arguments

Pull Request - State: closed - Opened by vondele about 4 years ago - 1 comment

#10 - Fix warnings on clang

Pull Request - State: closed - Opened by glinscott about 4 years ago

#9 - Add some additional arguments and output

Pull Request - State: closed - Opened by vondele about 4 years ago - 2 comments

#8 - Remove the non-sparse/batch datasets

Pull Request - State: closed - Opened by glinscott about 4 years ago

#7 - Add multithreaded tensor preparation on C++ side and binpack loader.

Pull Request - State: closed - Opened by Sopel97 about 4 years ago - 15 comments

#6 - Add experimental loss functions.

Pull Request - State: closed - Opened by Sopel97 about 4 years ago

#5 - Fix incorrect ctypes pointer binding

Pull Request - State: closed - Opened by Sopel97 about 4 years ago

#4 - Fix inverted `us` value from cpp data loader.

Pull Request - State: closed - Opened by Sopel97 about 4 years ago

#3 - Fix data loader doing flip instead of rotate180 for orient in halfkp.

Pull Request - State: closed - Opened by Sopel97 about 4 years ago

#2 - Clipped relu support

Pull Request - State: closed - Opened by glinscott about 4 years ago

#1 - [WIP] Defer data preparation to native code. Use sparse input tensors.

Pull Request - State: closed - Opened by Sopel97 about 4 years ago - 24 comments