They implemented wasm in transformer with cpu decoding for the wasm part.

the API is a very simple one where you slice parts off an initial Rect. the only feature it provides is that it tracks (x, y, h, w) for you.

it doesn't work well with intrinsic sizes - it's more of a top down, fixed size thing.

Background:

I've been looking at how to create a recurrent seq-to-seq model, that's not transformers. The ideas I implement do not work. It seems like off the well trodden path, there are traps everywhere - how should I tune parameters, add biases, normalize, is this dataset impossible, gradient explosion and vanishing, etc.

From a "research = gradient descent" point of view, I'm stuck at a point with no gradient - I have no idea what I'm doing wrong, or what to will get a better result. Am I missing a workflow. intuition, or tools, or other things?

Comments URL: https://news.ycombinator.com/item?id=36111233

Points: 4

# Comments: 2

I see the issue is that iff you explicitly marked it as inout, it would limit the code from a logic unification point of view. I guess this is already explored in prolog, which I have little experience with - but it feels like it would inherently limit the ability to scale a codebase to huge sizes. Maybe a combination of convention and naming would solve the problem.

https://imgur.com/a/7l377qr

simple logo of software engineer octopus using green yellow blue and orange databases

https://imgur.com/a/V0WqyFl

That there is a first element seems natural, however why should this first element to care about the 100010847238384392938474934th element?

I think if we pushed housing prices down, goods and services prices would increase.

That might be a good thing though.

Not sure about economic effect though, I'm not a electrical economic engineer haha

But the environment does matter to me, purely because I selfishly want to live longer and in a better environment.

One stumbling block would be addressability without TURN.

For ideas - remove ideas by banning / censorship, or add ideas by promoting 'better' ideas - this seems like a possible avenue to improving society's mental health. The main objection is whoever controls what to ban / allow is given a lot of power, so how do we decide these things.

For people - I think it's fine to be intolerant individually, however our institutions should not be intolerant. For example I can hire a cleaner for my house and refuse certain people. However it should not be possible for a company to refuse certain people as cleaners. I'm not sure how to formalize this into a law though.

Physical or mental violence because of intolerance of people and ideas are bad because I don't want violence on me. However I am not certain how to define mental violence.

If you want to post offensive material to some people/governments you need to work hard to stay anonymous, like the reporter example where they recommend using a public eg library place to connect.

If not you individually, maybe should select certain group of people to represent the people and create rules to enforce bad behavior is stopped?

Maybe we can call this body a government and call the rules laws.