Yixin Liu

Hi Griffin, I have also found this comparison very interesting! My guess is adjusting the likelihood has a more direct impact on the decoding output than adjusting the latent representation,...

Hi Tanya, I'm really sorry for the late reply. I've uploaded the checkpoint (as a generator) for NYT [here](https://drive.google.com/file/d/1O_XXQJLO7cgAmS0GPueV9SJtLQ8MHX7F/view?usp=sharing). I hope it's not too late :) I wanted to note...

Hi, could you provide more details on your finding (e.g. how did you notice this example? how did it affect your experiments?) I spot-checked around 10 examples and didn't find...

Hi, @aidejieceng, @thangld201, thanks a lot for bringing this to my attention! I found that there indeed is a misalignment between the tokenized input articles and reference summaries for CNN/DailyMail...

Hi, - Is the file (without the suffix --- ".tokenized" ) should be filled with the original sentence? Yes. - Which tokenizer should be used to tokenize the sentence in...

Hi! A checkpoint will be stored in a folder in the `./cache` directory along with the config.txt after the first evaluation step (the default setting should be 1000 update steps)....

Please check out here https://github.com/yixinL7/BRIO/tree/main/examples/raw_data (our new work). It contains the example files.

Good questions :) > Considering the Neural model's amazing capacity for memorization, the candidate generation of training set for evaluation model should be nearly perfect. That's not exactly true because...

I'd like to emphasize my point that **if the model is overfitting too much on the training set it would not perform well on the evaluation set**. So it's possible...

Hi! We didn't use the default version (roberta-based) of BertScore instead we used the bert-based model mainly because of different tokenizations. It has been a while so I couldn't recall...