alihan/LLMsPracticalGuide
1
0
Fork 0
mirror of https://github.com/Mooler0410/LLMsPracticalGuide.git synced 2023-08-15 01:42:19 +03:00
Code Issues Packages Projects Releases Wiki Activity

add tableformer as reference of tableQA example

Browse Source
This commit is contained in:
Jingfeng Yang
2023-04-30 01:04:47 -07:00
parent 80af08ef0d
commit fe35f3f345
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
awesome_examples/tableQA.md
View File

@@ -4,7 +4,7 @@ In this lesson, we ask the model to answer questions based on a table. The table
Comparing the following two examples, ChatGPT is vulnerable to table row order perturbation, while GPT4 is robust to table row order perturbation. Such robustness could probably be due to two reasons. The first reason is larger model size and more pretraining data of GPT4. Secondly, better truthfulness stemming from better RLHF alignment could help GPT4 follow different formats of the same instructions better. Comparing the following two examples, ChatGPT is vulnerable to table row order perturbation, while GPT4 is robust to table row order perturbation. Such robustness could probably be due to two reasons. The first reason is larger model size and more pretraining data of GPT4. Secondly, better truthfulness stemming from better RLHF alignment could help GPT4 follow different formats of the same instructions better.
Note that smaller finetuned models heavily suffers from such non-robustness issue, according to the paper: [TableFormer: Robust Transformer Modeling for Table-Text Encoding](https://arxiv.org/pdf/2203.00274.pdf) Note that smaller finetuned models heavily suffer from such non-robustness issue, according to the paper: [TableFormer: Robust Transformer Modeling for Table-Text Encoding](https://arxiv.org/pdf/2203.00274.pdf)
# Example 1 (2022/04/29) # Example 1 (2022/04/29)