RuntimeError: The size of tensor a (144) must match the size of tensor b (80) at non-singleton dimension 1
Closed this issue · 15 comments
Getting error with vocoder...
c = (c - self.stats["mean"]) / self.stats["scale"]
RuntimeError: The size of tensor a (144) must match the size of tensor b (80) at non-singleton dimension 1
I'm using pretrained models
pwg_TXHC/checkpoint-400000steps.pkl
Hi @MohanThippeswamy, can you tell me which recipe you are running?
Ah, I guess you are running the egs/l2-arctic/lsc recipe, and I think I know what the problem is... please give me some time to solve it. Sorry for this inconvenience!
you are right its egs/l2-arctic/lsc recipe.
Appreciate your turn around on the comments
Thanks
Ah, I guess you are running the
egs/l2-arctic/lscrecipe, and I think I know what the problem is... please give me some time to solve it. Sorry for this inconvenience!
@unilight, after using latest fixes the problem still remains the same (RuntimeError: The size of tensor a(144) must match the size of tensor b(80) at non-singleton dimension)
Hi @MohanThippeswamy can you tel me which step it is and maybe the whole log?
Hi @unilight, I was trying LSC inferencing using pretrained models
pretrained model related
pretrained_model_checkpoint=downloads/ljspeech_text_to_ppg_sxliu_aept/checkpoint-50000steps.pkl
npvc_checkpoint=downloads/s3prl-vc-ppg_sxliu/checkpoint-50000steps.pkl
npvc_name=ppg_sxliu
Please find log below
Stage 4: Network decoding
Decoding start. See the progress via exp/TXHC_bdl_1032_trial2_tm/results/checkpoint-50000steps/TXHC_dev/decode..log.
Decoding start. See the progress via exp/TXHC_bdl_1032_trial2_tm/results/checkpoint-50000steps/TXHC_eval/decode..log.
run.pl: 16 / 16 failed, log is in exp/TXHC_bdl_1032_trial2_tm/results/checkpoint-50000steps/TXHC_eval/decode..log
run.pl: 16 / 16 failed, log is in exp/TXHC_bdl_1032_trial2_tm/results/checkpoint-50000steps/TXHC_dev/decode..log
cont...
with open(config) as f:
FileNotFoundError: [Errno 2] No such file or directory: '/data/group1/z44476r/Experiments/s3prl-vc/egs/ljspeech/a2o_vc/exp/ppg_sxliu_taco2_ar/config.yml'
@MohanThippeswamy did you execute stages 0 to 3? The pre-trained models are used in the training, not for direct inference.
cont... with open(config) as f: FileNotFoundError: [Errno 2] No such file or directory: '/data/group1/z44476r/Experiments/s3prl-vc/egs/ljspeech/a2o_vc/exp/ppg_sxliu_taco2_ar/config.yml'
In my last fix, I added some instructions on modifying the paths in the config file. Please check.
@unilight, I executed stages 0to 2 but not 3.
your instruction from last fix is for conf/config.yml right? Bu anyway I have applied for all the models.
Pretrained models are not for direct inference. got it.
Wanted to use your pretrained models directly for inference where do I get those final models. please share the link that will help.
updating last fix (removed diifsinger module) is impact training..
ModuleNotFoundError: No module named 'seq2seq_vc.losses.diffsinger_l2_loss'.
So I'm struck at training so, I tried direct inferencing
@unilight,
any updates and further wanted to check on pretrained models?
Hi @MohanThippeswamy, sorry I'm a bit occupied these days. Will get back to you today!
@MohanThippeswamy See #36554fa
Cool!