Megatron NeMo GPT3

[VinodTipparaju]

I get several errors while running GPT3 with nvcr.io/nvidia/nemo:24.01.framework container. First I used the container as is, and I got an error "AttributeError: 'MegatronPretrainingSampler' object has no attribute 'batch_size'".

Then I did a git pull on all the repo's I could on /opt and now I get "TypeError: GPTDatasetConfig.init() got an unexpected keyword argument 'tokenizer'".

Could someone help?

Thanks,
Vinod.