You have a real nugget of an opportunity with this article. You might get another post to land with more traction by writing another article summarizing this one.
* Drop most of the pre-amble
* Explain your methodology, but drop most of the detail on each test.
* maybe a table summarizing the result
* consider focusing on the 2 or 3 most capable. Right now, it’s really hard to figure this out.
* Consider keeping some verbosity for the top models
I liked the detail and everything mentioned, so I wouldn't want OP to remove any of this. I was very captivated reading that. And I felt like I learned quite a bit.
One thing I do would like to finally see is a table for final comparison though.
But I enjoyed the description and process of how the OP went about testing. If that was removed, I might not have an idea how the tests were specifically done and may distrust the results. The description gave me an exact visual how OP went about all of this.
Because I don't think it's just about 2-3 most capable. I'm also very interested to see how models with different parameter sizes do and whether any have future potential to take over the most capable.
It's important to know what's the smallest open source model that can still be useful.
5 comments
[ 2.7 ms ] story [ 23.7 ms ] threadYou have a real nugget of an opportunity with this article. You might get another post to land with more traction by writing another article summarizing this one.
* Drop most of the pre-amble
* Explain your methodology, but drop most of the detail on each test.
* maybe a table summarizing the result
* consider focusing on the 2 or 3 most capable. Right now, it’s really hard to figure this out.
* Consider keeping some verbosity for the top models
One thing I do would like to finally see is a table for final comparison though.
But I enjoyed the description and process of how the OP went about testing. If that was removed, I might not have an idea how the tests were specifically done and may distrust the results. The description gave me an exact visual how OP went about all of this.
Because I don't think it's just about 2-3 most capable. I'm also very interested to see how models with different parameter sizes do and whether any have future potential to take over the most capable.
It's important to know what's the smallest open source model that can still be useful.
I could optimize more for content farming (and the result might be less self-indulgent and more readable) but then I would have a lot less fun.
I do appreciate the advice though.