^{Thanks for the model card image, Myscell!}

Qwen2.5-32B-Marigold-v0

Hasnonname's golden hands lay claim to another victim

Severian's notes: Good char/scenario portrayal and adherence. Speech patterns and formatting is followed well if you use chatml, while being slightly cursed, dramatic and in some cases refreshing when used with Mistral V7. Testers report user impersonation even on chatml which I never ran into, suspecting Qwen-isms and my prompt as the reason why I didn't ran into it as much. Testers also report good resulting prose from having thinking enabled. Not as varied across rerolls compared to both Mulleins even if it can be a little unhinged, and there's contention between which prose is better between this and Mullein v1, but I personally would use this as a daily driver.

72B coming soon.

Recommended settings

Context/instruct template: Chatml for the base experience, Mistral V7 shouldn't work and results to some system token bleed, but the prose can be dramatic and expressive with it too, sometimes showcasing better vocabulary and choice of words, try at your own risk.

Samplers: temperature at 0.9, min_p at 0.05, top_a at 0.3, TFS at 0.75, repetition_penalty at 1.03, DRY if you have access to it.

A virt-io derivative prompt worked best during our testing, but feel free to use what you like.

Thank you!

Big thanks to the folks in the trashpanda-org discord for testing and sending over some logs!

(datasets to be attributed later here)

Reviews

Got an impersonation, but reroll fixed that easily. Really like the prose so far, it's descriptive, detailed and lively, I'd say. It's by no means perfect, but its imperfections can be fixed easily by editing some parts for sure. But then again, r'really like its prose.

— Raihan

ngl you cooked with this one guys, especially with the thinking. I don't even wanna hide it anymore, it adds to the RP.

— AIELO

I can say that this new one is good at internal monologues and mental description. but less on action I suppose. Love it so far, nice descriptions of turmoil. It's better than R1 for me. If it had more reasoning/thinking it would be just chef's kiss perfect.

— Carmenta

[this model] is a mixed bag, there's some phrases that are eh here and there, but it's not that bad. I find the thinking more interesting as it seems to reinforce the card to be more in-character.

— Sam

Absolutely enjoy the dialogue. Follows the character card and does reasoning very well. In character thinking rather than 'bot' thinking. Peak gooning material.

— Mooth