Vily1998 commited on
Commit
a7c0616
1 Parent(s): 0eb243b

first commit

Browse files
._.gitattributes DELETED
Binary file (4.1 kB)
 
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ assert/truthfulqa_result.png filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -4,10 +4,9 @@ license: gpl-3.0
4
 
5
  # TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful Space
6
 
7
- [Shaolei Zhang](), Tian Yu, Yang Feng*
8
 
9
-
10
- TruthX is an inference-time method to elicit the truthfulness of LLMs by editing their internal representations in truthful space, thereby mitigating the hallucinations of LLMs. On the TruthfulQA benchmark, TruthX yields an average enhancement of 20% in truthfulness across various LLMs.
11
 
12
  This repository provides TruthX models trained on a variety of LLMs:
13
  - Llama-1-7B, Alpaca-7B
@@ -16,4 +15,9 @@ This repository provides TruthX models trained on a variety of LLMs:
16
  - Baichuan2-7B-Base, Baichuan2-7B-Chat
17
  - Chatglm3-6B-Base, Chatglm3-6B
18
 
 
 
 
 
 
19
  Please refer to [GitHub repo](https://github.com/ictnlp/TruthX) for specific usage scripts.
 
4
 
5
  # TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful Space
6
 
7
+ > [Shaolei Zhang](https://zhangshaolei1998.github.io/), Tian Yu, [Yang Feng](https://people.ucas.edu.cn/~yangfeng?language=en)*
8
 
9
+ TruthX is an inference-time method to elicit the truthfulness of LLMs by editing their internal representations in truthful space, thereby mitigating the hallucinations of LLMs. On the TruthfulQA benchmark, TruthX yields an average enhancement of 20% in truthfulness across various LLMs.
 
10
 
11
  This repository provides TruthX models trained on a variety of LLMs:
12
  - Llama-1-7B, Alpaca-7B
 
15
  - Baichuan2-7B-Base, Baichuan2-7B-Chat
16
  - Chatglm3-6B-Base, Chatglm3-6B
17
 
18
+ ## Results on TruthfulQA benchmark
19
+ - MC1 accuracy on TruthfulQA benchmark. More results refer to the paper.
20
+
21
+ ![truthfulqa_result](assert/truthfulqa_result.png)
22
+
23
  Please refer to [GitHub repo](https://github.com/ictnlp/TruthX) for specific usage scripts.
assert/truthfulqa_result.png ADDED

Git LFS Details

  • SHA256: 5d36cf716fd370b95f1c29c0fb1aec7313ca35d081e96fdb94b7506c03f8da6c
  • Pointer size: 132 Bytes
  • Size of remote file: 1.79 MB