Mysterious behavior
Hello.
The response to a long instruct is somehow stranger than v5 or v3. However, I have no idea what the logic is behind the weirdness in this merge configuration.🤯
It's a chat model, so it's fine as long as the conversation is understandable...
My knowledge of the LLM area is so poor that I'm not even sure which one is from which manufacturer, but you might want to try merging Nemo.
That guy's relatives have an unusually high command comprehension.
Well, anyway, thanks for the maintenance.
At some point I might try to make my own original merges using other models, and with any luck we might have something better than what I've made so far. Starcannon v3 was originally made by Auri using Celeste 1.9 and an older version of Magnum whereas v5 uses Magnum v2.5 with a heavier weighting towards Celeste.
I see...Celeste 1.9!
That guy was a model with interesting reactions, but he also rarely listened to commands.
I don't think the Magnum series is as command-oriented (neither of which is a problem, given the original direction of the model. It's just a matter of how it fits the use case), but he managed to neutralize Celeste with Magnum...😅
With the merging model of image generation AI, it's not always the case that versioning will yield better output for everyone. The same is probably true of LLM.
I think we should just think of each version of the model as a separate individual, a personality.
At some point I might try to make my own original merges using other models
I can only give you a taste of the model, but I'm looking forward to it!