yzhangcs/parser

'con-crf-roberta-en' won't load

dslim23 opened this issue · 5 comments

dslim23 commented
  • Python version: 3.10
  • Operating System: 13.3.1

Description

I was trying to load a model using Parser.load('con-crf-roberta-en') and predict on a sentence, but I encountered a ValueError saying "unknown url type: 'con-crf-roberta-en".

This error occurred running your code snippet in README.md:

from supar import Parser
 
parser = Parser.load('con-crf-roberta-en')

dataset = parser.predict('I saw Sarah with a telescope.', lang='en', prob=True, verbose=False)
dslim23 commented

upon a bit of investigation, it seems like the models here http://hlt.suda.edu.cn/~yzhang/supar/v1.1.0/ptb.crf.con.roberta.zip aren't accessible (perhaps due to server overload)

yzhangcs commented

@dslim23 Hi, you might need to install supar from source code

pip install -U git+https://github.com/yzhangcs/parser
dslim23 commented

I got it working with your suggestion thank you! is this still an issue #118 (comment)? I would like to use the pretrained weights. if so, what's the last version that could use these pt Roberta weights?

yzhangcs commented

@dslim23 Hi, I've submitted some patches, so installing from source are ok for RoBERTa :-)

Quang-elec44 commented

@yzhangcs Hi, can you check the again since I coudn't download this checkpoint?