Louis Fortier-Dubois

Hi @wcshds I haven't had time like I thought last week and then I was abroad for several days. I'm sorry I said I was gonna look at it last...

@wcshds I took your word that repeat was the bottleneck in wgpu. This made a lot of sense because we relied on the default implementation which launches as many `slice_assign`...

> Thank you, @louisfd for your research. This helped me go another route and leave nn modules as they're. In my PR (#1085), I am using custom deserializer with an...

This looks like the kind of bug that occurs because some methods always use the default device instead of the actual one... Default device stuff is under heavy refactoring in...

Fixed in #1054

> @louisfd Finally got the chance to start on #944. I'm going down the long route and trying to add support at large for `random()` on int tensors. When you...

> @nathanielsimard In `burn-wgpu/src/kernel/prng/bernoulli.rs`, it looks like there's already a `cast_elem` function, so maybe `convert_elem` or something similar? > > @louisfd Also, for the test that's failing, is there a...

Hi @allenqm Although you tagged Nathaniel, I think I can answer in his stead: - It's great that you made a default implementation for cumprod using log and exp 👍...

`cumprod` has also been asked for twice on discord: https://discord.com/channels/1038839012602941528/1038839013735399547/1234609825619644477 https://discord.com/channels/1038839012602941528/1091796857996451942/1202670351322972302 ``` rust let x = Tensor::arange(1..7 as i64, &device); x.cumsum(0) // [ 1, 2, 6, 24, 120, 720] ```

Hi Tom, I do believe fourier transform would be a nice addition to burn 😄 I made a branch so we can work on this together: https://github.com/tracel-ai/burn/tree/feat/ops/fft I looked at...