cerebras.modelzoo.common.utils.model.lora.disable_lora_merge_weights#

cerebras.modelzoo.common.utils.model.lora.disable_lora_merge_weights(lora_params_dict: Union[dict, List[dict]])[source]#

Sets merge_weights=False in LoRA parameters. This is helpful during eval mode to ensure that the weights don’t get folded prior to checkpoint loading.