Issues filed for karpathy/char-rnn

View Full Project
Do you use char-rnn? Leave a review!

Rate of open issues in the last 60 days

char-rnn open issues (View Closed Issues)
  • about 2 years Using multiple neuron layers > 3 results in NaN
  • about 2 years Installing Torch on Virtual Machine
  • about 2 years Training and Validation not handled correctly
  • over 2 years How to install this repository as a lua package
  • over 2 years 'vocab_attr' (a nil value) Error?
  • over 2 years Is it possible to grade style similarity?
  • over 2 years question about splitTable in model/LSTM.lua
  • over 2 years how to understand the parameters of LSTM.lstm constructor
  • over 2 years Can you please share the parameters values used for Linux data in the blogpost
  • over 2 years CR line terminators affecting training
  • over 2 years No Luarocks module found for util.OneHot
  • over 2 years Torch Issue
  • over 2 years what can be some possible applications of character level language modeling ?? out of curosity
  • over 2 years Question: Single line seq_length?
  • over 2 years Question about the training loss and validation loss.
  • over 2 years How to test the char-rnn model for some random text file? How to compile/run in ubuntu to get the next predicted character?
  • almost 3 years Multinomial Error
  • almost 3 years Endless "performing test c_has_sse4_1_2"
  • almost 3 years Subsequence Probability
  • almost 3 years Question About train.lua
  • almost 3 years memory problems and frequent crashes
  • almost 3 years Training keeps running out of memory in the "cloning rnn" stage
  • about 3 years Cannot Train: attempt to call field 'ClassNLLCriterion_updateOutput' (a nil value)
  • about 3 years Feature: beam search for improving global quality of new text samples
  • about 3 years code for init_from in training has few bugs
  • about 3 years Incorrect description of sampling temperature in Docs
  • about 3 years Add a check to see if training loss has gone to infinity
  • about 3 years Adjust learning rate based on score improvement
  • about 3 years allow the validation to be a seperate file
  • about 3 years add the ablity vary the temperature based on "scale"
  • about 3 years priming from an external file when sampling
  • about 3 years english and french text together causes crash
  • about 3 years How to use this code to model word-level RNN.
  • about 3 years Error using pre-trained model
  • about 3 years Are pre-trained models available anywhere?
  • about 3 years CPU is not being used 100%
  • about 3 years Memory Issue in Scaling sequence length
  • about 3 years bad argument #3 to 'ClassNLLCriterion_updateOutput' (torch.CudaTensor expected, got number)
  • over 3 years Out of memory error while evaluating a split
  • over 3 years Food for thought: Train data one-by-one?
  • over 3 years Problem with Torch
  • over 3 years Error: table index is nil
  • over 3 years luajit: Out of memory
  • over 3 years Scoring Functionality
  • over 3 years Unable to sample checkpoint trained with OpenCL
  • over 3 years Non-text data in input.txt
  • over 3 years Training with many short examples of variable length
  • over 3 years init_from fails with any given checkpoint
  • over 3 years Utf-8 support
  • over 3 years Should init_from parameter start train.lua from 1 if multiple checkpoints exist?
  • over 3 years train_loss and grad/param norm always are nan
  • over 3 years Training w/ GPU, sampling without GPU
  • over 3 years error in sample.lua: bad argument #2 to '?' (invalid multinomial distribution (sum of probabilities <= 0) at /root/torch/pkg/torch/lib/TH/generic/THTensorRandom.c:109)
  • over 3 years sample.lua fails to run: error in function addmm()
  • over 3 years input.txt max size
  • over 3 years Training determinism
  • over 3 years Word Level Encodings

char-rnn closed issues

  • over 2 years cunn and cutorch are not found (but they ARE installed)
  • over 2 years Can I delete older checkpoints?
  • over 2 years Loss is exploding, aborting
  • over 2 years usual prediciton error
  • over 2 years How to use a trained char-based RNN language model?
  • over 2 years Justification of cloning `rnn` and `criterion`
  • almost 3 years time/batch about same regardless of gpuid
  • almost 3 years on clone_many_times()
  • almost 3 years For info, started on a version of char-rnn using Element Research rnn modules
  • almost 3 years char-rnn is 7% slower on NVIDIA 940M since commit 82baee
  • almost 3 years rnn and ocr
  • about 3 years learningRate not saved
  • about 3 years Why the lstm is not built using the container?
  • about 3 years bad argument #1 to 'set'
  • about 3 years In the sampling, should we subtract the max first, for numerical stability?
  • about 3 years Question: How to print out the activations of non-linear states and gates?
  • about 3 years Why the seq_length and batch_size are set 50 or other value?
  • about 3 years openLC error
  • about 3 years Question: How to reset all states after each sequence? [Duplicate]
  • about 3 years Hints for a Unified Religion Text Generator
  • about 3 years After the last commit
  • about 3 years Why do we need to carry over the rnn_state in the evaluation for batch prediction?
  • about 3 years Simple recipe to have another 6% computation speedup in CUDA mode
  • about 3 years Plausible model misuse
  • about 3 years checkpoints not saving after upgrade to gpu capability
  • about 3 years second RMSprop update yields NaN loss?
  • about 3 years Warning: comparison of integers of different signs
  • about 3 years Gradient divided by sequence length
  • about 3 years About GPU Usage and Monitoring
  • about 3 years Question: why don't we learn the initial states?
  • about 3 years How is the weighted matrix trained?
  • about 3 years Where I sholud put the download codes?
  • about 3 years newline character in prime text
  • about 3 years params_lstm:copy(x) does not work for GPU version LSTM RNN
  • over 3 years Global var 'path' is null in utils
  • over 3 years Only copy the state of top_h?
  • over 3 years Training the net locks up the operating system
  • over 3 years How do I know GPU is running instead of CPU?
  • over 3 years Peephole connections
  • over 3 years cutorch problem - invalid argument at (...)/cutorch/lib/THC/THCTensor.cu:32
  • over 3 years Question: why doesnt state get reset after each epoch?
  • over 3 years Multi-GPU support
  • over 3 years Train Error on GPU ClassNLLCriterion.lua:34: bad argument #1 (field weights does not exist)
  • over 3 years Sampling to file
  • over 3 years can we update char-rnn model ?
  • over 3 years CUDA support broken against current torch/nn
  • over 3 years Gradient magnitute depends on seq_length
  • over 3 years Error loading module 'libnn' while trying to run