ViBES-Face / expert_checkpoint.json

Switch to Expert-1-only (motion) checkpoint: 19.95GB -> 0.86GB; loaders merge with GLM-4-Voice base (bit-identical)

dfbcaf5 verified 1 day ago

263 Bytes

	{
	"expert": "motion",
	"format": "expert1_only",
	"base": "glm-4-voice-9b",
	"note": "Contains ONLY the trained motion expert (Expert 1). Expert 0 (text/audio) == the GLM-4-Voice base; reconstruct + merge via training/expert_io.py:load_expert1_checkpoint."
	}