spunkedy @spunkedy

Joined November 2008

Tweets

2K
Followers

133
Following

176
Likes

216

spunkedy @spunkedy

a year ago

@josevalim One of the good talks at black hat last year was on this type of attack vector for privilege escalation

0 0 0 406 1

View Details

@josevalim @VKatsuba Right now a lot of the models have been trained really well on node frameworks. So full stack like next js can be leveraged faster. As we transition to being able to be force multiplied with AI we absolutely have to rethink our ways of working. Have to build a bridge sometimes.

0 0 0 30 0

View Details

spunkedy @spunkedy

2 years ago

@EddieValeJr @SteveSkojec They were here in Texas just just a fyi

0 0 1 34 0

View Details

spunkedy @spunkedy

2 years ago

@bcardarella I have been using Aider with examples of read only to generate some localized help for this basically.

0 0 0 43 0

View Details

spunkedy @spunkedy

2 years ago

@akoutmos Love it

0 0 1 18 0

View Details

spunkedy @spunkedy

2 years ago

@s_igu @jskalc @elixirphoenix I use this one too behind a Cloudflare proxy

0 0 1 44 0

View Details

spunkedy @spunkedy

2 years ago

I am genuinely impressed at the stark contrast of customer service of between US amazon.com and Spain amazon.es. Never had an issue working with US. Come to spain, 10+ issues with delivery in a month. Chat agents hanging up. Maybe culture? @AmazonHelp

1 0 1 50 0

View Details

spunkedy @spunkedy

3 years ago

@josevalim Yeah. Because of the deploy feature of creating derivative works essentially creates an overlay I could see it getting murky. Having an exception clause for those would probably work

0 0 0 28 0

View Details

spunkedy @spunkedy

3 years ago

@josevalim AGPL could potentially be argued that the live books developed on it would mean that companies would have to open source that part. Hashicorp went towards BSL and tofu is now a thing. Ngnix / nginx plus might be a good pattern.

0 0 0 216 0

View Details

spunkedy @spunkedy

3 years ago

@sean_moriarity I have gone through a bunch of examples and one of the “classes” at elixir conf for ML. But do you know of a good resource for implementing a new model architecture specifically for mixtral? I see most of the conversations use Bert. Python seems to use AutoLM.

0 0 0 38 0

View Details

spunkedy @spunkedy

3 years ago

@AndrewBeckUSA Been in Austin a while. Next time if you want I can make some recommendations for places to eat.

0 0 0 119 0

View Details

spunkedy @spunkedy

3 years ago

@bcardarella

0 0 0 34 0

View Details

spunkedy @spunkedy

4 years ago

@bcardarella Lol. Looks like the support is already done. I wonder if I can get it building locally until it gets merged

1 0 0 18 0

View Details

spunkedy @spunkedy

4 years ago

@bcardarella Sweet. I will see if I can help. Otherwise maybe the phoenix channels are working good enough now. 4 years ago it wasn't quite there

1 0 0 19 0

View Details

spunkedy @spunkedy

4 years ago

@sean_moriarity dreamfusion3d.github.io Python implementation github.com/ashawkey/stabl… I think it's a bit out of scope for bumblebee. Does elixir clustering for batching of models work ? If not that might be a nice feature