# Online training with Julia

**URL:** <https://discuss.bayesflow.org/t/online-training-with-julia/95>\
**Category:** General\
**Created:** [July 8, 2024, 1:34pm UTC](https://discuss.bayesflow.org/t/online-training-with-julia/95 "2024-07-08T13:34:03Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![kiante](https://yyz1.discourse-cdn.com/flex007/user_avatar/discuss.bayesflow.org/kiante/32/67_2.png) [@kiante](https://discuss.bayesflow.org/u/kiante)\
**Post date:** [July 8, 2024, 1:34pm UTC](https://discuss.bayesflow.org/t/online-training-with-julia/95/1 "2024-07-08T13:34:03Z")

</div>

Rather than using Numba (or other methods in Python), has anyone attempted to use Julia to do online training (I’ve done so offline)? The implementation I am thinking of would involve something like [PythonCall](https://github.com/JuliaPy/PythonCall.jl). I am curious if speed benchmarks tests already exist here.

---

<div class="post-metadata">

**Author:** ![KLDivergence](https://yyz1.discourse-cdn.com/flex007/user_avatar/discuss.bayesflow.org/kldivergence/32/15_2.png) [@KLDivergence](https://discuss.bayesflow.org/u/KLDivergence)\
**Post date:** [July 9, 2024, 8:39am UTC](https://discuss.bayesflow.org/t/online-training-with-julia/95/2 "2024-07-09T08:39:31Z")

</div>

@elseml Has some experience with integrating Julia simulators, but I believe so far only in an offline context?

---

<div class="post-metadata">

**Author:** ![elseml](https://yyz1.discourse-cdn.com/flex007/user_avatar/discuss.bayesflow.org/elseml/32/19_2.png) [@elseml](https://discuss.bayesflow.org/u/elseml)\
**Post date:** [July 9, 2024, 9:47am UTC](https://discuss.bayesflow.org/t/online-training-with-julia/95/3 "2024-07-09T09:47:12Z")

</div>

Hey everybody, I have indeed also only used it in an offline context so far. But it would be very interesting to hear about your experiences if you decide to try it out @kiante.

---

<div class="post-metadata">

**Author:** ![nw95](https://avatars.discourse-cdn.com/v4/letter/n/edb3f5/32.png) [@nw95](https://discuss.bayesflow.org/u/nw95)\
**Post date:** [August 10, 2024, 8:28pm UTC](https://discuss.bayesflow.org/t/online-training-with-julia/95/4 "2024-08-10T20:28:38Z")

</div>

I am using online-training with Julia (as in, the heavy lifting of the generative model is done in Julia). I have seen some performance improvement but the main constraint is and was the training of the neural networks. I sadly didn’t do any “formal” benchmark testing.

One big advantage (for me) is that I can easily swap out different models.