# How to retrieve gradient value in custom Flux training loop?
**URL:** https://discourse.julialang.org/t/how-to-retrieve-gradient-value-in-custom-flux-training-loop/82435
**Category:** Machine Learning
**Tags:** question
**Created:** [June 8, 2022, 1:48pm UTC](https://discourse.julialang.org/t/how-to-retrieve-gradient-value-in-custom-flux-training-loop/82435 "2022-06-08T13:48:44Z")
**Posts on this page:** 4
**Page:** 1
### Author:  [@BatyLeo](https://discourse.julialang.org/u/BatyLeo)
#### Post date: [June 8, 2022, 1:48pm UTC](https://discourse.julialang.org/t/how-to-retrieve-gradient-value-in-custom-flux-training-loop/82435/1 "2022-06-08T13:48:44Z")
Hello,
I have a few questions about custom training loops in Flux. According to the [documentation](https://fluxml.ai/Flux.jl/v0.10/training/training/#Custom-Training-loops-1), I should write something like that :
```julia
function my_custom_train!(loss, ps, data, opt)
ps = Params(ps)
for d in data
gs = gradient(ps) do
training_loss = loss(d...)
# Insert what ever code you want here that needs Training loss, e.g. logging
return training_loss
end
# insert what ever code you want here that needs gradient
# E.g. logging with TensorBoardLogger.jl as histogram so you can see if it is becoming huge
update!(opt, ps, gs)
# Here you might like to check validation set accuracy, and break out to do early stopping
end
end
```
1. If the `ps` argument is `Flux.params(my_model)` like in the `Flux.train!` method, is the `ps = Params(ps)` row redundant ?
2. `gs` seems to be an instance of `Zygote.Grads`, how can I retrieve gradient value in the loop ? I tried `gs[ps]` and `gs[my_model]` without any success.
---
### Author:  [@albheim](https://discourse.julialang.org/u/albheim)
#### Post date: [June 8, 2022, 2:00pm UTC](https://discourse.julialang.org/t/how-to-retrieve-gradient-value-in-custom-flux-training-loop/82435/2 "2022-06-08T14:00:31Z")
1. Yes, it shouldn’t be needed.
2. The `Grads` type contains the fields `grads` and `params`, so you can access them as `gs.grads`.
---
### Author:  [@CarloLucibello](https://discourse.julialang.org/u/CarloLucibello)
#### Post date: [June 8, 2022, 2:02pm UTC](https://discourse.julialang.org/t/how-to-retrieve-gradient-value-in-custom-flux-training-loop/82435/3 "2022-06-08T14:02:15Z")
> 1. If the `ps` argument is `Flux.params(my_model)` like in the `Flux.train!` method, is the `ps = Params(ps)` row redundant ?
yes
> 1. `gs` seems to be an instance of `Zygote.Grads`, how can I retrieve gradient value in the loop ? I tried `gs[ps]` and `gs[my_model]` without any success.
```julia
for p in ps
print(gs[p])
end
```
---
### Author:  [@BatyLeo](https://discourse.julialang.org/u/BatyLeo)
#### Post date: [June 8, 2022, 2:16pm UTC](https://discourse.julialang.org/t/how-to-retrieve-gradient-value-in-custom-flux-training-loop/82435/4 "2022-06-08T14:16:42Z")
Thank you for your help ! This is exactly what I was looking for.