issues
search
Equationliu
/
Kangaroo
Implementation of Kangaroo: Lossless Self-Speculative Decoding via Double Early Exiting
https://arxiv.org/abs/2404.18911
39
stars
5
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Kangaroo when bsz is greater than 1.
#6
cool-xiang
closed
1 week ago
2
Encountering NaN output at a specific batch ID every run, and no change observed upon adjusting the learning rate
#5
Zerohclmax
closed
1 month ago
0
a question
#4
cool-xiang
closed
2 months ago
2
In line 263 of train.py， predict = model(inputs_embeds=data["hidden_states_early"]
#3
cool-xiang
closed
2 months ago
1
Training procedure of Kangaroo.
#2
tim-pan
closed
2 months ago
5
why warmup when evaluating
#1
EganGu
closed
4 months ago
2