lantiga
Re:Show HN: A fully open-source (Apache 2.0)implementation of llama
on Tuesday March 28, 2023 @04:07PM (#35347073)
Correct, the original is GPL 3.
To produce this implementation from the LLaMA paper we started from github.com/karpathy/nanoGPT, the LLaMA architecture is really similar to GPT. For instance we added rotary positional encoding starting from the original RoPE repo published with the paper.
We finally ran the original model to make sure the two models were numerically.
Re:Show HN: A fully open-source (Apache 2.0)implementation of llama
on Tuesday March 28, 2023 @03:52PM (#35346863)
Some form of that is very likely the future
Re:Show HN: A fully open-source (Apache 2.0)implementation of llama
on Tuesday March 28, 2023 @03:51PM (#35346845)
The AI ecosystem is almost entirely Apache 2/MIT/BSD, and GPL is just incompatible with it.
This is a blocker to mixing and matching, a simple Apache 2 rewrite fixes that problem.
Weights? It’s another issue but we’ll be looking forward to fixing that too.
Re:Deep Learning with PyTorch
on Friday November 22, 2019 @02:41AM (#21603630)
Pretty much (co-author here). You can approach the book as a developer with little or no background in ML.
The purpose is exactly to get developers up to speed so that they can start perusing the already excellent online docs.
Hope you like it :-)
Re:Ki – A lisp for your JavaScript
on Tuesday June 17, 2014 @09:19AM (#7904215)
Shen is pretty amazing, it's a full fledged functional language with lots of features and lots of compilation backends. I don't think there's a lot of ground for a comparison with ki, which is much more limited in scope.
Re:Ki – A lisp for your JavaScript
on Tuesday June 17, 2014 @06:41AM (#7903626)
I have ideas around providing CSP in ki, but so far I haven't implemented them. All there is in ki right now is the (letc ...) form (see http://ki-lang.org/api.html#misc), which provides very similar functionality to LiveScript's backcalls. It is no CSP but it allows to straighten up nested async calls.
Re:Ki – A lisp for your JavaScript
on Tuesday June 17, 2014 @06:36AM (#7903616)
It's actually named KL (with a lower-case L), but yes, that was close.
Re:Ki – A lisp for your JavaScript
on Tuesday June 17, 2014 @06:35AM (#7903612)
The main difference is structural sharing: I can update a Clojure map with a new value, which will create another map that shares the vast majority of the data with the former map. With plain JavaScript, you'd have to clone your object, update it and freeze it again. In addition Clojure/ClojureScript/mori offer a whole API designed to leverage immutable data structures.
Re:Ki – A lisp for your JavaScript
on Monday June 16, 2014 @11:43AM (#7899630)
I agree, LispyScript is very neat, closer to a full-fledged language than ki. On the other hand, ki is focused on ClojureScript's immutable data structures and related idioms - in fact it's just a thin layer around those.
Re:Ki – A lisp for your JavaScript
on Monday June 16, 2014 @08:13AM (#7898583)
Glad you find it interesting, looking forward to future feedback.
Re:Ki – A lisp for your JavaScript
on Monday June 16, 2014 @08:08AM (#7898568)
[Hi, author here] Yes, that is basically the idea. ki functions are js functions, same goes for vars, which makes it easy to jump from js to ki and back, without giving up on immutability, etc. Note that ki is not really intended as a ClojureScript alternative at all. It's more a way to take advantage of its data structures and some of its expressiveness from a pure JS ecosystem.