Fianna Lin

ValueError, subprocess.CalledProcessError, 训练失败

2

用推荐的PAI-DSW跑的，已经将./facechain/train_text_to_image_lora.py的1033行```train_loss += avg_loss.item() / args.gradient_accumulation_steps```后面加上了梯度，但依然不成功。 打开链接成功，上传图片成功，但是训练失败，报错信息如下： ``` 2024-03-19 16:35:16,756 - modelscope - INFO - Use user-specified model revision: v2.0 {'clip_sample_range', 'rescale_betas_zero_snr', 'variance_type', 'thresholding', 'dynamic_thresholding_ratio', 'sample_max_value'} was not found in config....

File prepare_dataset/scannet/prerocess_scannet.py: 1. computes the `segid_to_pointid` map by reading _`vh_clean_2.0.010000.segs.json` 2. computes the `instance_segids` map by reading `.aggregation.json` 3. then derive the `pointid_label` relation (line93, `semantic_labels[pointids] = RAW2SCANNET[labels[i]]`) However, it...