Sao10K/Llama-3.3-70B-Vulpecula-r1

Text Generation • Updated 30 days ago • 1.33k • 36

Note Nearly unrecognizable as Llama except for some of its -isms. Very smart, "niche adult" knowledge is stellar, and it molds engaging responses without feeling contrived. Occasionally falls into repetition death spirals; a backend with DRY, or better parameter tuning, should sort that out (I find the default recommendations cause it to fall into a confusion spiral instead.) Probably the best RP reasoner I've used.

trashpanda-org/QwQ-32B-Snowdrop-v0

Text Generation • Updated 7 days ago • 1.39k • 66

Note Produces responses that are clever, witty, and away from the beaten path. The RP reasoner dream realized.

trashpanda-org/Llama3-24B-Mullein-v1

Updated Mar 5 • 1.9k • 7

Note Despite the name, this is an MS3! And a decent one at that. The name comes from the fact that in testing, it seems to work better with Llama3 formatting than Mistral V7. Nobody knows why. It wasn't in any of the data.

DatToad/Chuluun-Qwen2.5-72B-v0.08

Text Generation • Updated Jan 27 • 41 • 6

Note Chuluun plays unconventional and "otaku-coded" archetypes with aplomb. Toad struck gold very early on, and polished it to a shine.

d-rang-d/MS3-RP-Broth-24B

Text Generation • Updated Mar 3 • 73 • 6

Note Very involved 24B merge; possibly the most usable MS3 I've seen. Great for multiturn writing especially.

EVA-UNIT-01/EVA-Qwen2.5-72B-v0.2

Text Generation • Updated Dec 3, 2024 • 127 • 19

Note Sharp as a tack and very responsive to user input; a scene favourite for a reason.

EVA-UNIT-01/EVA-Qwen2.5-32B-v0.2

Text Generation • Updated Nov 14, 2024 • 268 • 54

Note While it's never going to be as good as the 72B, EVA 32B maintains the brunt of EVA 72B's intelligence while being usable on (high-end!) consumer hardware.