DaSiWa v1 is outdated
Can we get DaSiWa v4 / v4 Turbo gguf quants?
Do you mean for Minimax H3? Is it actually better? Can you link me to the correct best source? (I have not tested v4)
https://civitai.red/models/2877206/dasiwa-minimax-h3?modelVersionId=3391862
Yeah, minimax H3 dasiwa series. There's V1 in the dumps and there's a lot in there since. But Kobold cringes and doesn't eat .safetensors so I can't really compare it myself.
P.S. Oh, and the developer states that their Turbo subversions work in 4 steps.
P.P.S. There're also LoRAs on their main page https://civitai.red/user/Darksidewalker/models
Actually kobold does support safetensors. You just need to load the correct components.
Nonetheless I shall try to download and run the latest version
hi @Nesaliti I've converted it for you. Enjoy
Aye, this works, thank you very much! Though result promised by the author in 4 steps doesn't seem to make wonders. 6 seems fine.
The exact v4 listed file from the civitai page was used, from int8 format. Perhaps they had a different finetune or lora?
The exact v4 listed file from the civitai page was used, from int8 format. Perhaps they had a different finetune or lora?
Yeah, I sifted through author page a bit. Here is the LoRA for it, as far as I understand: https://civitai.red/models/2929860/minimax-h3-turbo-or-ref2va-or-fl2va-or-hybrid
Works fine now.