Foundry_24B_V.4

Goal of this merge is to "upgrade" Cogwheel model to mistral-small 3.2, given how good new mistral is.

Model is very creative, with good narration abilities and "live" dialogues. It follows instructions fine, and stable enough. Tested context length was 12k, quality wasn't degrading much. (Probably good up to 16-22k context)

Model gives long enough replies, eos token works, most of time. Quality of text is good, no nonsense, but not boring. Bias of model is neutral, it will just work in given setting, be it utopia, grimdark, classic fantasy, sci-fi or erp. NSFW wise, model not censored, could use swears if prompted, and not prone to fall to erp instantly in non-erp scenarios.

Model prone to answer for user if user is descripted in char card, but not always. Model catches formatting patterns easily.

Ru also was tested, and it is good, even enough for light roleplay, just minor writing errors could occur. Ru erp wasn't tested.

Tested on ~ 400 replies, Mistral template, T 1.01,