alerem18

i can save model parameters using BSON in a hook, but i don't know how can i use the model and test it on the env(without training, only testing)

not working on julia 1.9

1

`[ Info: Precompiling GameZero [9da27670-f782-11e9-1da1-f53579315bfe] ERROR: LoadError: InitError: could not load library "C:\Users\Administrator\.julia\artifacts\fa03230282478d82c2295049d440075426a17202\bin\SDL2_ima ge.dll"`

New Bug

1

i tried the code twice and it was throwing an error in both also some times it stuck(when i'm not drawing objects, it stuck for long time), it's much slower...

Convolutional network slower than tensorflow on CPU

13

```julia using Flux using MLDatasets: MNIST, CIFAR10, CIFAR100 using Flux: logitcrossentropy, setup, Adam, train! using Flux.OneHotArrays: onehotbatch, onecold using Statistics: mean using Flux.MLUtils: DataLoader using ProgressBars: tqdm, set_postfix using Flux.Zygote:...

explicit differentiation for RNN gives wrong results

32

i tried to implement a RNN MODEL to classify Mnist Dataset but i get an accuracy around 40-50% even with running it for more than 20 epochs, while in pytorch,...

can't use masks in multi-head-attention layer

6

### Motivation and description let's say we have an array of shape (embedding_size, seq_len, batch_size), our padding mask will have a shape of (seq_len, batch_size) which can't be used in...

flax is significantly slower than pytorch

1

minist classification using convolutional networks link: https://flax.readthedocs.io/en/latest/quick_start.html same approach on pytorch is significantly faster, (almost 5 times faster) flax version takes 50 seconds per epoch!!!

Machine Translation With Transformers

3

can't run this example on jax or pytorch backend it just works on tensorflow backend https://keras.io/examples/nlp/neural_machine_translation_with_keras_nlp/ also inferencing is significantly slower than a similar implementation in pytorch, like 8 times...

as postgre supports arrays you should add types like string[] or integer[], as using those type currently throws an error in migration up

it's a good feature if you support socks proxies as i think currently you support only http/s proxies