Andrea Ferretti

Hi all, it's great that this PR landed in 0.12, so that we can experiment with various attention modification techniques. What is exactly the API for this feature, though? The...

Hi @evinpinar, thank you for your prompt response! So, one thing I gather from your example is that `set_attn_processor` accepts a dictionary mapping layer names to processors, and will use...

Hi @damian0815 , thank you, it is very useful to have another example to learn from!!

> I don't think it's possible to have a toSeq(a: typed) I think the proposal is to have various overloads of `toSeq` - one for iterators and the other ones...

I have the same issue. In fact, I am using CUDA 7.0. Any chance the project will be ported to more recent versions of CUDA? I tried doing this myself,...

@pierre-rouleau I just updated the tutorial to mention that is now included in the official documentation. If everything still works, it will show up in the [site](http://andreaferretti.github.io/factor-tutorial/) soon

¯\_(ツ)_/¯ Unfortunately, I don't remember how I set up the site, so I don't know how to refresh it

I am pretty sure there was a way in the settings to tell Github that the gh-pages branch should be built automatically using Jekyll and the markdown in the README...

See [this PR](https://github.com/andreaferretti/neo/pull/45). The issue is: `shallowCopy` was kind of needed to have matrices that share storage (think of taking a row, or the transpose of a matrix, without making...

It would amount to marking `data` as shallow in places like [here](https://github.com/andreaferretti/neo/blob/master/neo/dense.nim#L24) and [here](https://github.com/andreaferretti/neo/blob/master/neo/dense.nim#L31), and similarly in `cudadense` etc, then removing `shallowCopy` where it appears, running tests and checking what...