Okay, so I’ve been messing around with local LLMs lately, you know, trying to see what all the fuss is about. I decided to pit two popular choices against each other: la-lafayette and gw. It was kind of a spur-of-the-moment thing, really.
Setting Things Up
First, I had to get both of them up and running. I already had la-lafayette installed from a previous experiment, so that part was easy. Getting gw going took a bit more effort. I had to dig through some documentation and fiddle with a few settings, but I eventually got it working. Nothing too complicated, just your usual setup headaches.
The First Run
I started by feeding both models the same simple prompt, something like, “Write a short story about a cat.” I wanted to see how they handled basic creative tasks. La-lafayette churned out a decent little story, nothing spectacular, but it was coherent and followed the prompt.
Gw, on the other hand, gave me…well, it was a story, I guess. It was a bit more rambling and less focused. Let’s just say it was more “abstract.”
Digging Deeper
Then, I decided to throw something a bit more challenging at them. I gave them a more complex prompt, involving a specific setting, characters, and a plot twist. This is where things got interesting.
- la-lafayette:It did a better job of sticking to the details I provided. The story felt more structured and the characters were more consistent.
- gw:It still struggled a bit with coherence. It kind of went off on tangents and introduced elements that weren’t really relevant to the prompt.
Trying different prompts
I spent the whole afternoon doing different prompts on both of them.
- la-lafayette:It start to get a little bit boring, all the answer are similar structrue.
- gw:It give me a lot of different answers, some of them are really out of my mind!
My (Very Unscientific) Conclusion
So, after spending a good chunk of time playing around with both, here’s my completely subjective take: la-lafayette seems to be better at following instructions and producing coherent outputs, especially for more structured tasks. Gw, while sometimes a bit all over the place, feels more…unpredictable? It might be better for brainstorming or generating more “out there” ideas, and some times really surprise me.
Honestly, it probably depends on what you’re looking for. If you need something reliable and predictable, la-lafayette might be the way to go. If you’re feeling adventurous and want to see what kind of weirdness a model can come up with, give gw a shot. Both are fun to play with, in their own ways!
