radq

8 ms·

31.
by radq 3y ago
We were missing two architecture patterns that were needed to get deeper nets to converge: residual nets [1] which solved gradient propagation, and batch normalization [2] which solved initialization. [1] Residual nets (2015): https:/
32.
by radq 3y ago
I have a similar setup and decided to use Tailscale. It's been working seamlessly so far.
33.
by radq 3y ago
As someone that rides a bicycle around human-driven vehicles, I will take my chances with autonomous cars. The autonomous car isn't going to drive drunk, use a phone, or drive erratically because the driver is feeling frustrated.
34.
by radq 3y ago
I've been working on training models for this problem. Haven't shipped as a product yet because the inference cost is too high (takes 10-15 minutes to transcribe each song). Would love to try it out on the music you're trying
35.
by radq 3y ago
I'm having similar problems and haven't found any way to resolve it. Even had previously indexed pages start to drop off the index. Ended up putting that project on the back-burner.
36.
by radq 3y ago
Sounds like they don't have rate limiting implemented? That seems odd, and it's also surprising it isn't talked about in the post.
37.
by radq 3y ago
> I'm sure things are better now but I haven't messed with it since then. I had the same experience ~3 months ago. Gave up and switched to Nvidia 3090s for my workloads.
38.
by radq 3y ago
This reminds me of the Octuple MIDI tokenization scheme introduced in the MusicBERT paper ( https://arxiv.org/pdf/2106.05630.pdf ). Would be interesting to see how much of a performance difference results from using a sm
39.
by radq 3y ago
Mark Zuckerberg talks about this in their Q1 earnings call. "I think that there's an important distinction between the products we offer and a lot of the technical infrastructure, especially the software that we -- that we write t
40.
by radq 3y ago
I'm working on a chord detection model. It's still training but already gives decent results for some songs. Some examples: https://harmonote.com/songs/nfUjirHCnasBZ9qQrsbzM8-heat-wave... https://h
41.
by radq 3y ago
I started programming when I was 12, mostly writing small scripts to customize PHP forum software. I joined a number of online forums and mailing lists with more experienced folks doing similar things and learned a ton from reading their co
42.
by radq 3y ago
I've had the issue of generating random code after the completion with other models as well; it's due to how the models are trained. You need to stop generating when you encounter token(s) that indicate you're done - see htt
43.
by radq 3y ago
Agreed, it's starting to get really annoying when models that aren't get announced as being "open source".
44.
by radq 3y ago
It is most likely due to the text encoder - see "Character-Aware Models Improve Visual Text Rendering". https://arxiv.org/abs/2212.10562
45.
by radq 4y ago
The larger context length makes me think they have a more memory-efficient attention mechanism.
46.
by radq 6y ago
Not OP, but I switched to using Microsoft Office Lens.
47.
by radq 8y ago
Amazon Web Services - AWS IQ | SDE and/or Web Dev | Seattle, WA or Vancouver, BC | Onsite, Full-time AWS is cloud company. You've probably heard of us. Although it’s a large organization it’s made up of lots of small product teams
48.
by radq 9y ago
Amazon Web Services | SDE and/or Web Dev | Seattle, WA or Vancouver, BC | Onsite, Full-time AWS is cloud company. You've probably heard of us. Although it’s a large organization it’s made up of lots of small product teams. We are
49.
by radq 9y ago
What are the issues with it? Anything other than the size/complexity?
50.
by radq 11y ago
The proposal only applies to local variables with non-null initializers. I'm pretty sure you can't autowire local variables with Spring, and they don't have initializers anyway.
51.
by radq 11y ago
Lombok does rewrite bytecode at compile time only. Your dependencies will not have to depend on Lombok.
52.
by radq 12y ago
How does this compare to PredictionIO?
53.
by radq 12y ago
It did not forward me at least.
54.
by radq 12y ago
Our open source project has a significant social network component, one option could be to use it as a base and get rid of the stuff you don't need. Not sure how closely it fits your use-case. https://hummingbird.me/use
55.
by radq 12y ago
Just out of curiosity, why does Tarsnap not work for you?
56.
by radq 12y ago
What we do is run Varnish in front of S3 to cache files on machines with cheaper bandwidth. https://github.com/hummingbird-me/ansible-deploy/blob/master... It is very little work to set up, and you don't
57.
by radq 12y ago
"They" can be and is commonly used as a gender-neutral singular. http://en.wikipedia.org/wiki/Singular_they
58.
by radq 12y ago
We open-sourced our startup ( https://github.com/hummingbird-me/hummingbird ) and I think it was fairly successful. Personally I feel that the git history is useful so I would prefer not to release just a snapshot. Would
59.
by radq 12y ago
I did a project along these times some time back, training a Hidden Markov Model on accelerometer + gyroscope data sampled every 100ms gave basically 100% classification accuracy. Can share the code with you if you are interested.
60.
by radq 12y ago
Looks like Rails, Angular and Pusher to me.

More ›