tang-ed

Results 15 comments of tang-ed

bow_loss下不去，不知道怎么回事

bow_loss稳定在2左右，一直下不去

bow_loss下不去，不知道怎么回事

lm_loss都下到0.001左右了

It seems decoder() in SelfTokenizer is wrong

With the update iteration of the version, some tokenizers are no longer used. My transformers version is 4.11.0

There are no Chinese training dataset in this project.

You can find the data set yourself. I think the open source open domain chat data set of Tsinghua University is good.

There are no Chinese training dataset in this project.

xxxxxxxx xxxxxxxxxx xxxxxxxxxx xxxxx xxxxxx xxxxxxxx xxxxx xxxxxxx xxxxxxxxxx xxxxx .........

an error when run python train.py

Did you change the code? If so, please send it to have a look. Because of the original code, I run well. It should also be noted that tokenizer is...

an error when run python train.py

That shouldn't be. My guess is that there is a problem with your programming environment, because I didn't see the complete error report. You can try to run it in...

loss==nan

Thank you for your use. This project has been updated. You can try again to see if there is a similar situation.

[BUG]: /bin/bash: line 0: export: `=/usr/bin/supervisord': not a valid identifier Error: failed to run torchrun --nproc_per_node=1 --nnodes=1 --node_rank=0 --rdzv_backend=c10d --rdzv_endpoint=127.0.0.1:29500 --rdzv_id=colossalai-default-job train.py --use_trainer on 127.0.0.1, is localhost: True, exception: Encountered a bad command exit code!

This is what I ran on a paid platform using Resnet as a case study

[BUG]: /bin/bash: line 0: export: `=/usr/bin/supervisord': not a valid identifier Error: failed to run torchrun --nproc_per_node=1 --nnodes=1 --node_rank=0 --rdzv_backend=c10d --rdzv_endpoint=127.0.0.1:29500 --rdzv_id=colossalai-default-job train.py --use_trainer on 127.0.0.1, is localhost: True, exception: Encountered a bad command exit code!

I am able to run any pytorch program normally, except for the example provided by Colossal AI