Longxu Dou

@margauxbo @Carbonhell @exitxingling @lwmlyy Thanks for your interests in multispider! Please refer to [code](https://github.com/longxudou/multispider) and [data](https://huggingface.co/datasets/dreamerdeo/multispider) for further discussion.

Hi @srewai , many thanks for your interest in unisar. To make the unisar work on your own dataset, the ONLY job is to call the step2 to generate the...

@srewai Actually, mBART shows a promising (but not good as English) performance in German (i.e., question and DB are in German and the output SQL also involves German headers.) In...

@srewai You can wait for our multilingual-Spider to train your own parser. It supports Chinese/Japanese/German/French/Spanish/Vietnamese. About 10K data and 120 tables. Q1: I will need to create sqlite db out...

Hi @srewai, this is the paper about multilingualSpider: https://arxiv.org/pdf/2212.13492.pdf . Welcome to read :) The codebase and dataset will be released in about one month. (I need to prepare more...

@epejhan90 Thanks for your interest! Q1: What are the preprocesses to add my table to db and be able to run three steps? The codebase doesn't support this function. But...

@epejhan90 This is because the released checkpoint is no-value version, which is trained by "query_toks_no_value" rather than "query_toks" in Spider dataset. If you want to fix this (make SQL contains...

@epejhan90 I think you could first fine-tune our non-value checkpoint to be "valuable" with construting the training corpus. Otherwise, it's time-consuming to adopt one GPU to fine-tune the BART-large model...

@bekhzod-olimov Hi, thanks for your interests! Please refer to [code](https://github.com/longxudou/multispider) and [data](https://huggingface.co/datasets/dreamerdeo/multispider) for further discussion.

Thanks for the excellent codebase and instructions! Eagerly awaiting a fix for the OOM issue.