Skip to content

Spider results higher than paper. #16

@jacklishufan

Description

@jacklishufan

Hi
It seems the pretrained checkpoint has execution accuracy of 98.4 on Spider test set. Is this desired? It is significantly higher than table 1 (70.0).

The full result is

                     easy                 medium               hard                 extra                all                 
count                470                  857                  463                  357                  2147                
=====================   EXECUTION ACCURACY     =====================
execution            0.996                0.974                0.983                0.994                0.984   

The result is the same for CLLM and Pretrained LLM

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions