How to call after training lora weights using train_lcm_distill_lora_sd_wds.py

[yangzhenyu6]

I use the following script to get the training results, but I get an error when using Dreamshaper7 as a pre-trained model, I also specified Dreamshaper7 as the teacher model when training.:
#!/bin/bash

Define the variables

PRETRAINED_TEACHER_MODEL="/ai/yzy/latent-consistency-model-main/LCM_Dreamshaper_v7"
OUTPUT_DIR="/ai/yzy/latent-consistency-model-main/output"
RESOLUTION=512
LORA_RANK=64
LEARNING_RATE=1e-6
LOSS_TYPE='huber'
ADAM_WEIGHT_DECAY=0.0
MAX_TRAIN_SAMPLES=200000
DATALOADER_NUM_WORKERS=4
TRAIN_SHARDS_PATH_OR_URL='/ai/yzy/latent-consistency-model-main/dataset.tar'
VALIDATION_STEPS=50
CHECKPOINTING_STEPS=50
CHECKPOINTS_TOTAL_LIMIT=10
TRAIN_BATCH_SIZE=8
GRADIENT_ACCUMULATION_STEPS=1
SEED=453645634

Run the training script

python ./LCM_Training_Script/consistency_distillation/train_lcm_distill_lora_sd_wds.py
--pretrained_teacher_model=$PRETRAINED_TEACHER_MODEL
--output_dir=$OUTPUT_DIR
--mixed_precision=fp16
--resolution=$RESOLUTION
--lora_rank=$LORA_RANK
--learning_rate=$LEARNING_RATE
--loss_type=$LOSS_TYPE
--adam_weight_decay=$ADAM_WEIGHT_DECAY
--max_train_samples=$MAX_TRAIN_SAMPLES
--dataloader_num_workers=$DATALOADER_NUM_WORKERS
--train_shards_path_or_url=$TRAIN_SHARDS_PATH_OR_URL
--validation_steps=$VALIDATION_STEPS
--checkpointing_steps=$CHECKPOINTING_STEPS
--checkpoints_total_limit=$CHECKPOINTS_TOTAL_LIMIT
--train_batch_size=$TRAIN_BATCH_SIZE
--gradient_checkpointing
--enable_xformers_memory_efficient_attention
--gradient_accumulation_steps=$GRADIENT_ACCUMULATION_STEPS
--use_8bit_adam
--resume_from_checkpoint=latest
--num_train_epochs=10
--seed=$SEED