Could you merge this model?
Hey Undi,
Not sure if you've seen froggerics creativity benchmark but your PsyMedRP model scored rather well on it!
https://huggingface.co/datasets/froggeric/creativity
Notes on your model from the benchmark:
Undi95/PsyMedRP-v1-20B:
"Great writing with lots of details, taking sufficient time to develop the plot. The small context size though is a limiting factor for consistency."
I use this model a lot and it's creative wordplay and 'soulful' responses are incredible. I was wondering if you could do a merge of this model
with CohereForAI/c4ai-command-r-v01: https://huggingface.co/CohereForAI/c4ai-command-r-v01
c4ai-command-r-v01 scored 2nd place on the creativity benchmark. It follows instructions extremely well, has long context and is very smart but it's just missing that creative spark that and amazing wordplay that PsyMedRP-v1-20B has.
I think a fusion of these two model would be extremely fruitful. Let me know what you think! :)
Hello, I sadly can't merge these two because the architecture is not the same.
One is Llama, the other is Cohere.
Hello, I sadly can't merge these two because the architecture is not the same.
On is Llama, the other is Cohere.
Oh that's a shame.
Do you have know of any other models that have been merged with PsyMedRP that you or someone else has created? I'd love to try it :)
This model is what got me into model merging. I'd gently gestured toward my Psyonic-Cetacean-20B if you want try something similar but different. Its got a bit more of a novelwriting and adventure bias, and like PsyMed takes up a large amount of VRAM due to the way stacked merges work, but I've been using it since I cooked it basically uninterrupted aside from experimental models.
https://huggingface.co/jebcarter/psyonic-cetacean-20B
https://huggingface.co/mradermacher/psyonic-cetacean-20B-i1-GGUF
This model is what got me into model merging. I'd gently gestured toward my Psyonic-Cetacean-20B if you want try something similar but different. Its got a bit more of a novelwriting and adventure bias, and like PsyMed takes up a large amount of VRAM due to the way stacked merges work, but I've been using it since I cooked it basically uninterrupted aside from experimental models.
https://huggingface.co/jebcarter/psyonic-cetacean-20B
https://huggingface.co/mradermacher/psyonic-cetacean-20B-i1-GGUF
Hey Jeb, thankyou. I've actually downloaded your model a while back actually and liked it a lot. It's quite eloquent and writes a solid story. :) The thing I'm really looking for in a model though is imagination and creative inspiration, I find PsyMedRP (although a bit unstable, has a short context and can go off the rails at times ) has a REALLY VIVID imagination! Some of the scifi scenes, characters, worlds and ideas it creates in my mind are astounding. I mainly use LLM's for creative inspiration and this model comes up with things that are really out of this world. I have yet to find a model to match it.