Clojurians Log v2

I'm so embarrassed -- clearly I didn't grep the whole source. Thank you!!

No need to be embarrased (!) I went to this page and did ctrl-f "missing": https://techascent.github.io/tech.ml.dataset/ I knew about col/set-missing but your question was a good one, made me curious if there were other ways too. 🙂

First thought:

user> (require '[tech.v3.dataset :as ds])
nil
user> (def ds (ds/->dataset {:a (range 10)}))
#'user/ds
user> (require '[tech.v3.dataset.column :as col])
nil
user> (assoc ds :a (col/set-missing (:a ds) [1 3 5 7 9]))
_unnamed [10 1]:

| :a |
|---:|
|  0 |
|    |
|  2 |
|    |
|  4 |
|    |
|  6 |
|    |
|  8 |
|    |

Hi @zengxh! https://neanderthal.uncomplicate.org/, and the related https://github.com/uncomplicate/deep-diamond, are high-performance Clojure libraries that can replace pytorch in some of what it does. If you care about performance, then the coding style with these libraries will be slightly different from typical Clojure, since careful memory management is needed. There has been recent interest in learning these libraries (among other things) at the https://scicloj.github.io/docs/community/groups/real-world-data/ study group, with a couple of relevant talks in recent meetings. All of that is discussed more a the https://scicloj.github.io/docs/community/chat/#the-clojurians-zulip.

The version in the video is: https://github.com/karpathy/nanoGPT There are several other implementations that might be easier to port across: • https://github.com/karpathy/minGPT • https://github.com/keyvank/femtoGPT (RUST) • https://github.com/jaymody/picoGPT • https://github.com/benisalla/micro-tiny-gpt-from-scratch • https://jaykmody.com/blog/gpt-from-scratch/ Lllama • https://github.com/Lightning-AI/lit-llama • https://github.com/akanyaani/miniLLAMA What are your goals? • Inference only or training/fine tuning? • Simple or fast?

I did this https://github.com/rthadani/andrej needs documentation and stopped somewhere in the 2nd part of the series. I was hoping to change this to use clay and then write the doc but I’ve hade trouble with my editor

@rohit_ your current draft looks really interesting. How was your experience using pytorch through interop? Did it raise any trouble or inconvenience?

i tried and lost it when pytorch kicks in.