Add unconditional lora support so Ideogram 4 inference will more closely resemble the full pipeline results. I pushed a finetuned unconditional lora to the hub as an adapter.

toolkit/config_modules.py

@@ -620,6 +620,9 @@ class ModelConfig:
         # mainly for decompression loras for distilled models
         self.assistant_lora_path = kwargs.get('assistant_lora_path', None)
         self.inference_lora_path = kwargs.get('inference_lora_path', None)
+        # a lora that stays inactive except during the unconditional (negative)
+        # CFG pass -- used to learn the unconditional branch without a second model
+        self.unconditional_lora_path = kwargs.get('unconditional_lora_path', None)
         self.latent_space_version = kwargs.get('latent_space_version', None)
 
         # only for SDXL models for now