Rui Meng

@sbmaruf I just notice that (1)`interleave_datasets` only samples indices once and reuse for all epochs, and (2) it's limited by the smallest dataset. Do you figure out an alternative way...

@lhoestq I encountered another related issue. I use load_dataset() for my json data and set_transform() for preprocessing. But it hangs at the end of the epoch if `dataloader_num_workers>=1`. It appears...

I use my own summarization data and modified a bit the default [`config-transformer-base-1GPU.yml`](https://github.com/OpenNMT/OpenNMT-py/blob/master/config/config-transformer-base-1GPU.yml). Now I think I found the key, it's still about the copy generator (I reported a similar...

No, same error occurs as long as copy_attn is enabled, with pytorch 1.2.

I updated apex to nightly and pytorch to 1.3. With model_dtype=fp16 and apex_opt_level=O2, this problem seems to have disappeared. But I did run into `CUDA error: misaligned address` once at...

@hzahera Apologies. I don't think the new code is perfectly compatible with the previous checkpoints. The parameters should be reloadable but some configs could be buggy. We are already working...

Hi @smolPixel , we are developing a new package and some demos on the basis of huggingface transformer. But no doubt it will not work for previous RNN models. I...

Please stay tuned as we are preparing for a tutorial at ECIR. A new [codebase](https://github.com/midas-research/dlkp) will be released and it can support fine-tuning better. It supports more pretrained models and...

Facing the same problem on A100. Attempts of reducing batch size doesn't seem to help since there are many documents in the index. It looks like an OOM, no problem...

> The instructions in that notice do not work in the docker image. > > ``` > FROM nvidia/cuda:11.4.0-runtime-ubuntu20.04 > ... > RUN apt-key del 7fa2af80 > ADD https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-keyring_1.0-1_all.deb ....