I can't wait for you to release code that uses AMRBART！

OSError: Can't load tokenizer for 'bert-base-uncased'. If you were trying to load it from 'https://huggingface.co/models', make sure you don't have a local directory with the same name. Otherwise, make sure 'bert-base-uncased' is the correct path to a directory containing all relevant files for a BertTokenizer tokenizer.

I have downloaded bert-base-uncased, but I always prompt that I can't find it. And I changed the default of load_model_dir to the path of the file.

Dear author, thanks to the amazing work and open-sourced code.
I'm puzzled by some of your treatment of the Parser AMR:

1. The first question is about the transition amr. First, I downloaded the "transition-amr-parser" folder and the "amr_general" folder, as mentioned in TSAR. Then, using the command "pip install transition-amr-parser" to download the required transition amr. However, when I run the "python amrparse.py" command, I get errors "no module of fairseq" and "ImportError: urllib3 v2.0 only supports OpenSSL 1.1.1+". For the above two errors, I run the command "pip install fairseq==0.10.2" and the command "pip install urllib3==1.26.15" respectively. However, when I run the "python amrparse.py" command, I still get "KeyError: 'stack_transformer_6x6_nopos' ", which occurs in the "amrparse.py" file with the line "parser = AMRParser.from_checkpoint(path)", seems to be related to the fairseq package. Has the author ever encountered this problem?

2. The second question is about AMRBART. May I ask if the data needs to be compressed after being parsed into AMR by AMRBART?

Thank u for your patience.

Thank you for sharing! However, the results I obtained on the RAMS dataset are lower than the paper. I am wondering if this could be due to virtual environment configuration issues. While running the script you provided, I encountered an error "TypeError: init() got an unexpected keyword argument 'label_smoothing'", which might be caused by a version mismatch in some package. Could you please provide me with details of your environment?
========for span F1========
dev

Precision: 40.9049 Recall: 50.4510 F1: 45.1792

test

Precision: 42.5606 Recall: 50.1748 F1: 46.0551
========for head F1========
dev

Precision: 46.7093 Recall: 57.6099 F1: 51.5901

test

Precision: 49.3327 Recall: 58.1585 F1: 53.3833

Is it possible to provide an example script of how to go from a single raw AMR graph to the simplified representation proposed in the paper?

Dear author,
Thank you for this work.
I found the paper well-researched, however, I have a question regarding a specific point. In Section-2.2-Missing spans, the concept of "match" was mentiond, and I would like to seek further explanation to better understand it.

If a generated span partially matches a node, we add a new node to represent this span and inherit connectives from the partially matched node. We also add a special edge between this node and the new node to indicate their overlap. If a generated span fails to match any existing nodes, we add a new node and connect it to the nearest nodes to its left and right with a special edge.

Or could you kindly provide guidance on which part of the code corresponds to the aforementioned steps?
Thank you very much for considering my inquiry.

Hi.
Thanks for your code! According to the paper, results using AMRBART on WikiEvents test set based on RoBERTa large is as bellow.
78.35 76.29 73.07 70.83

But I only received low performance as follow:

'========for Head F1 Identification========' 
Precision: 71.9361 Recall: 81.4889 F1: 76.4151
'========for COREF F1 Identification========' 
Precision: 70.6927 Recall: 80.0805 F1: 75.0943
'========for Head F1 Classification========' 
Precision: 67.1403 Recall: 76.0563 F1: 71.3208
'========for COREF F1 Classification========' 
Precision: 66.0746 Recall: 74.8491 F1: 70.1887

It puzzled me. I set the same hyperparameters according to the appendix and use the preprocessed data provided by you . Did I do something wrong?

Thanks for reply.