Is this the 'better than Lazarus' model?
" I hope you manage to find this model useful. I have an enormously better release than Laz coming. If you'd like, keep an eye out; it has no boundaries and is highly competent [33B-Epilogue is uploading...]."
Earlier this was teased on a Lazarus discussion by the author. Is this the model in question?
It is, it's now my go-to 30b for now. Model card is back-asswards as I made it public before adding that in ha. It's a complex merge based on best practices learned from previous experiments. Of course, results will be subjective for everyone.
It is, it's now my go-to 30b for now. Model card is back-asswards as I made it public before adding that in ha.
I suppose writing a full model card might take a while, but for the time being could you put down the prompt / instruction style?
Alpaca style prompting, its a merge of the following: [SuperCOT[superhot2-8k[(wizardlmuncensored[StoryV2(hippogriff+manticore)])]
(Source: Digitous himself in DM's)
Alpaca style prompting, its a merge of the following: [SuperCOT[superhot2-8k[(wizardlmuncensored[StoryV2(hippogriff+manticore)])]
(Source: Digitous himself in DM's)
Thanks for answering. So it's a superhot 8k model too?
So I can't use alpha on this and should use scaling of 4?
Alpaca style prompting, its a merge of the following: [SuperCOT[superhot2-8k[(wizardlmuncensored[StoryV2(hippogriff+manticore)])]
(Source: Digitous himself in DM's)Thanks for answering. So it's a superhot 8k model too?
Apologies for the wait - Ah, it's based on the old prototype of SuperHot - so no 8K yet; just updated the model card. Thank you for your patience.