issues
search
voidism
/
DoLa
Official implementation for the paper "DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models"
https://arxiv.org/abs/2309.03883
419
stars
50
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Do you have the code for the visualization?
#19
JiaxinQin0814
opened
4 weeks ago
0
The way to calculate the log_probs
#18
laurenlong
opened
4 months ago
0
data for gpt-judge training
#17
itsmemala
closed
4 months ago
1
Typo? mode = "early_exit_contrastive" -> mode = "dola-static"
#16
young-j-park
closed
5 months ago
1
How to install the environment if I don't have root permission?
#15
hummingbird2030
opened
6 months ago
0
Adding support for T5 models
#14
itshuey
closed
6 months ago
0
Query regarding JS Divergence mean over batches.
#13
talha1503
closed
6 months ago
2
What tool do you use to get the token prediction of each layer of large language models for Figure 2?
#12
frankdarkluo
closed
7 months ago
3
Isn't the range of the JS divergence [0,1]? Why are there values greater than 1 in Figure 2?
#10
Jeryi-Sun
closed
8 months ago
1
Should apply model.norm layer to hidden_states[early_exit_layer] ?
#9
githubhyz
opened
11 months ago
1
fix bug for contrastive sample decoding
#8
HillZhang1999
closed
8 months ago
0
concerns about your experiments of TruthfulQA-MC
#7
0-KaiKai-0
closed
8 months ago
2
Some questions about the idea of the paper
#6
Rh-Dang
closed
6 months ago
1
Figure 2 subtoken "_"
#5
githubhyz
closed
1 year ago
0
Support for LLaMA-2
#4
ayyyq
closed
8 months ago
15
questions about code
#3
Cheung-Z
closed
8 months ago
1
dola可以适配xverse模型吗?
#2
xxm1668
closed
7 months ago
1
Fix typo in README.md
#1
eltociear
opened
1 year ago
0