Experimental KTO tuned model. Should work well for RP or instruct. Use chatML format. Will provide detailed instructions later.