this post was submitted on 23 Nov 2023

1 points (100.0% liked)

LocalLLaMA

3 readers

1 users here now

Community to discuss about Llama, the family of large language models created by Meta AI.

founded 1 year ago

MODERATORS

communick@poweruser.forum

What is your top 3 fav LLM's (with settings) for creative storytelling? (alien.top)

submitted 11 months ago by LeanderGem@alien.top to c/localllama@poweruser.forum

16 comments fedilink hide all child comments

Just wondering what are peoples go to LLM's when they really want to get creative/unique stories. What I am mainly looking for is creative sparks of brilliance that really feel like it's something fresh/unique/creative with writing that just flows well.

Mine are as follows (using KoboldCPP):

1.) psymedrp-v1-20b.Q5_K_M Settings: Quick Preset: Mayday, Temp 1.2, Top P Sampling 0.99

link: https://huggingface.co/TheBloke/PsyMedRP-v1-20B-GGUF

Notes: This one really gets creative and has lines that make me smile for how outlandish, bizzare and creative they sometimes are. Sometimes goes a little off the deep end but generally this is one is a powerhouse of creativity.

2.) dolphin-2.1-mistral-7b.Q8_0 Settings: Quick Preset: Mayday, Temp 1.2, Top P Sampling 0.95

link: https://huggingface.co/TheBloke/dolphin-2.1-mistral-7B-GGUF

Notes: Love this little creative maestro. Very eloquent and the wordplay is fun and fresh with occasionally very poetic/flowery prose (sometimes a tad too much). Overall very solid.

3.) collectivecognition-v1.1-mistral-7b.Q8_0 Settings: Quick Preset: Mayday, Temp 1.3, Top P Sampling 0.95

link: https://huggingface.co/TheBloke/CollectiveCognition-v1.1-Mistral-7B-GGUF

Notes: Also just as creative as dolphin.2.1 with possibly a more serious tone. Also on occasion uses very techy words like ' algebra ', 'algorithmic' when describing mundane things but overall a solid creative 7b.

So what are your guys faves? (with settings too please).

top 16 comments

sorted by: hot top controversial new old

[–] brobruh211@alien.top 1 points 11 months ago (1 children)

Glad to see dolphin 2.1 still getting love! I too prefer it to even its newer versions when it comes to creative storytelling.

Have you tried any 34Bs? I found even just the Q3_K_M quant of Nous-Cabypara 34B to be more creative and coherent than any smaller model I've tried. Try using a low-temp preset like Kobold's Liminal Drift. If on SillyTavern, I found the Roleplay context template and instruct mode to work well with this.

[–] LeanderGem@alien.top 1 points 11 months ago

Haven't tried that one actually, thanks for the heads up, off to download now. Yeah I also like the older Dolphin to the newer one myself, the newere one feels a little more 'muffled' in a sense.

[–] Dry-Judgment4242@alien.top 1 points 11 months ago (2 children)

Lzlv 70b still the best model by a mile for story writing.

[–] _supert_@alien.top 1 points 11 months ago (1 children)

It is very good. Synthia and Euryale are also very good.

For 13Bs Mythomax, Chrono-Hermes and Tiefighter are cool.

[–] LeanderGem@alien.top 1 points 11 months ago

Thanks. Another one that I've been testing out and is showing great potential is mythomist-7b

https://huggingface.co/TheBloke/MythoMist-7B-GGUF

[–] ChiefBigFeather@alien.top 1 points 11 months ago (1 children)

Lzlv 70b is good, but I found Airo 70b 3.1.2 a little better at grasping context.

[–] LeanderGem@alien.top 1 points 11 months ago

The little guys just can't beat the bigboys it seems. Time to upgrade, lol.

[–] Postorganic666@alien.top 1 points 11 months ago

Goliath is great. But it requires some GPU lol

[–] mcmoose1900@alien.top 1 points 11 months ago (1 children)

Tess and Capybara 34B are both excellent, I'm just running them merged. There's really nothing else with such long context, and that is so smart at handling the long context.

[–] LeanderGem@alien.top 1 points 11 months ago (1 children)

Nice, thanks. Is this the one you are referring to?:

https://huggingface.co/TheBloke/Capybara-Tess-Yi-34B-200K-GGUF

[–] mcmoose1900@alien.top 1 points 11 months ago

Yeah. I am running the exl2 quant though, it lets you fit a ton of context.

Tess 1.1 by itself is also said to be good: https://huggingface.co/models?sort=modified&search=tess+1.1

[–] AnomalyNexus@alien.top 1 points 11 months ago (1 children)

Athena V4

Think it's aimed at ERP but remarkably pleasant at general upbeat female AI persona.

dolphin.2.1 with possibly a more serious tone.

Tried the yi dolphin one a bit...seems to provide much shorter & curt responses. Def doesn't feel story telling like to me. Maybe the mistral version is better

[–] LeanderGem@alien.top 1 points 11 months ago

Agreed, I tried Yi-Dolphin too, it just didn't have the same charisma as dolphin 2.1 in my opinion.

[–] BlueIdoru@alien.top 1 points 11 months ago (1 children)

I'm currently using Emerhyst 20b.

[–] LeanderGem@alien.top 1 points 11 months ago

Downloaded this one a while ago, will be fun to see how it does.

[–] oodelay@alien.top 1 points 11 months ago

Still loving 33b wizardLM for smut but Tiefighter 13b is becoming my go-to