LM-BFF
LM-BFF copied to clipboard
princeton-nlp
[ACL 2021] LM-BFF: Better Few-shot Fine-tuning of Language Models https://arxiv.org/abs/2012.15723
The function returns `label_ids`. However, when `print_name in ['sst-5', 'mr', 'cr', 'mpqa', 'subj', 'trec']`, `label_ids` is not defined. I guess there should be `return labels` in this branch?
`run.py` accepts parameter `num_k`, which should control how many training examples per class we have. However, it's not used anywhere in the code. It seems that `num_sample` is used instead.
Hi! Really thanks for your work. And I have a question about prediction at inference with demonstrations. As the paper mentioned, the final prediction logits is by averaging the the...
Hi, when I use your code to auto-generate labels, the label pairings are empty. The image is the details. And the output file is also empty. But I still have...
The ``truncate_head`` param of ``tokenize_multipart_input`` function is not enabled, when call it for normal prompt tuning (i.e., w/o demonstration). In fact, for some extreme long dataset, we have to manually...
I am trying to apply LM-BFF to some sentiment analysis datasets and Named Entity Recognition in Arabic. I am having difficulties in creating the templates as the order in which...
I am attempting to identify the most suitable prompt for a word sense disambiguation task using the WiC dataset. A single data point from this dataset could look like this:...
Now, by trying the command TAG=exp TYPE=prompt-demo TASK=SST-2 BS=2 LR=1e-5 SEED=42 MODEL=roberta-large bash run_experiment.sh which is in the README I got this error: Traceback (most recent call last): File "/LM-BFF/run.py",...
have been trying to install the requirements file in a virtual environment, and I am encountering a problem with dependencies. It says there may be an issue with the numpy...
Metadata
Owner
Metadata
[ACL 2021] LM-BFF: Better Few-shot Fine-tuning of Language Models https://arxiv.org/abs/2012.15723