Renqian Luo

@dukebw Hi Brendan, thanks for your response. Yes, I just found that the derived final model has to be run from scratch and is not implemented yet. And I also...

@dukebw If it works for M = 10, then I think it would be better to set M = 10. BTW, I run your latest code and the training ppl...

@zihaozhang9 train_search.py里的model就是要搜索的网络 nao包含的就是encoder-predictor-decoder结构 model的valid accuracy作为评价该model的指标，encoder-predictor要学的就是从model到它的valid accuracy这的这种映射关系。

label是attention map中attentiond的对角线集中率，公式可以见paper里。PTB数据的处理方式和其他相关工作（DARTS）一致

It can constrain the model to generate within specific words. However we do not actually turns it on in all the example tasks currently.

Hi, you can format the task to a sequence-to-sequence format which is more convenient for the generative model. For NER task, you can follow the format in this [paper](https://arxiv.org/pdf/2106.03598.pdf) We...

If you run the scripts, the `src` is passed to the program via the `--user-dir` parameter, which contains the `language_modeling_prompt` task. If you use it in our own code, please...

Hi @VisionaryMind , this is due to a rename bug. We have fixed it now. Please pull the latest code and re-download the QA-PubMedQA-BioGPT-Large.tgz checkpoint

pull the latest code and try this: ``` m = TransformerLanguageModel.from_pretrained( "checkpoints/Pre-trained-BioGPT-Large", "checkpoint.pt", "data/BioGPT-Large", tokenizer='moses', bpe='fastbpe', bpe_codes="data/BioGPT-Large/bpecodes", min_len=100, max_len_b=1024) ```

Yes, if you run it from path other than the root path of BioGPT, you should add it to the PYTHONPATH