problems encountered in the process of using the HippoSSKernel function of the S4 Module

[Ssw2001]

hi,Excuse me for disturbing you. When I was running the S4 module, an error occurred when it came to the nplr function of the HippoSSKernel function, as shown below.When I remove this module, everything runs smoothly, so I suspect there might be an issue within the S4 module itself. Is this problem due to incorrect parameter settings or is it incompatible with my PyTorch version 3.7.10? I am very much looking forward to your reply as this is very important to me. Thank you!

CUDA extension for cauchy multiplication not found. Install by going to extensions/cauchy/ and running python setup.py install. This should speed up end-to-end training by 10-50% Falling back on slow Cauchy kernel. Install at least one of pykeops or the CUDA extension for efficiency. {'diffusion_config': {'T': 200, 'beta_0': 0.0001, 'beta_T': 0.02}, 'wavenet_config': {'in_channels': 1, 'out_channels': 1, 'num_res_layers': 36, 'res_channels': 256, 'skip_channels': 256, 'diffusion_step_embed_dim_in': 128, 'diffusion_step_embed_dim_mid': 512, 'diffusion_step_embed_dim_out': 512, 's4_lmax': 1000, 's4_d_state': 64, 's4_dropout': 0.0, 's4_bidirectional': 1, 's4_layernorm': 1, 'label_embed_dim': 128, 'label_embed_classes': 71}, 'train_config': {'output_directory': '/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/sssd_label_cond', 'ckpt_iter': 'max', 'iters_per_ckpt': 400, 'iters_per_logging': 10, 'n_iters': 10000, 'learning_rate': 0.0002, 'batch_size': 6}, 'trainset_config': {'segment_length': 1000, 'sampling_rate': 100, 'finetune_dataset': 'ptbxl_all', 'data_path': '/home/sunsw/sunshiwen/work/SSSD-ECG/src/ptb_xl/processed-data/test_ptbxl_1000.npy'}, 'gen_config': {'output_directory': '/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/sssd_label_cond', 'ckpt_path': '/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/sssd_label_cond/'}} output directory /home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/sssd_label_cond/ch256_T200_betaT0.02_train Traceback (most recent call last): File "mytrain1.py", line 212, in train(train_config) File "mytrain1.py", line 53, in train net = mySSD_ECG1(model_config).cuda() File "/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/models/mySSD_ECG1.py", line 258, in init label_embed_dim=label_embed_dim if label_embed_classes > 0 else None) File "/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/models/mySSD_ECG1.py", line 207, in init label_embed_dim=label_embed_dim)) File "/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/models/mySSD_ECG1.py", line 122, in init layer_norm=s4_layernorm) File "/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/models/S4Model.py", line 1190, in init bidirectional=bidirectional) File "/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/models/S4Model.py", line 1074, in init self.kernel = HippoSSKernel(self.h, N=self.n, L=l_max, channels=channels, verbose=verbose, **kernelargs) File "/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/models/S4Model.py", line 975, in init w, p, B, = nplr(measure, self.N, rank, dtype=dtype) File "/home/sunsw/sunshiwen/work/SSSD-ECG/src/sssd/models/S4Model.py", line 441, in nplr V_inv = V.conj().transpose(-1, -2) IndexError: Dimension out of range (expected to be in range of [-1, 0], but got -2)

[juanlopezcode]

Hi, firstly, it seems that you are in an incorrect repository, you are working with SSSD-ECG, not SSSD. Secondly, it seems that you have altered the network as you call now mySSD_ECG1, if that is the case, I'm unable to assist here. In any case, it seems that the problem comes from V parameter in S4 model, following up on what and where made it might help.

[Ssw2001]

Hi, firstly, it seems that you are in an incorrect repository, you are working with SSSD-ECG, not SSSD. Secondly, it seems that you have altered the network as you call now mySSD_ECG1, if that is the case, I'm unable to assist here. In any case, it seems that the problem comes from V parameter in S4 model, following up on what and where made it might help. ; Thank you very much for your answer, it has been of great help to me. After switching to a higher version of PyTorch, I was able to get it running. However, I would like to confirm that after I ran the command python setup.py install for the file /root/SSSD/src/extensions/cauchy/setup.py, a file named cauchy_mult.py appeared, and the content is:

def bootstrap(): global bootstrap, loader, file import sys, pkg_resources, importlib.util file = pkg_resources.resource_filename(name, 'cauchy_mult.cpython-38-x86_64-linux-gnu.so') loader = None; del bootstrap, loader spec = importlib.util.spec_from_file_location(name,file) mod = importlib.util.module_from_spec(spec) spec.loader.exec_module(mod) bootstrap()

Is the installation successful? After placing this file in /root/SSSD/src/extensions/cauchy, it worked for me. Is my approach reasonable?