1063 位用户此时在线
24小时点击排行 Top 10：
- 本站自动实时分享网络热点
- 24小时实时更新
- 所有言论不代表本站态度
- 欢迎对信息踊跃评论评分
- 评分越高，信息越新，排列越靠前
2
1
0
A new method was able to delete 40% of LLM layers with no drop in accuracy.
This makes them mich cheaper and faster.
The method combines pruning, quantization and PEFT.
They tested this across various open source models.
Each family of models had a maximum amount of layers…
时政
(
twitter.com
)
由
Lior⚡
提交
A new method was able to delete 40% of LLM layers with no drop in accuracy.
This makes them mich cheaper and faster.
The method combines pruning, quantization and PEFT.
They tested this across various open source models.
Each family of models had a maximum amount of layers that could be deleted before accuracy dropped:
- Mistral - 30%
- Lama 70B - 40%
- Llama 13B - 50%
Great paper.
点击图片查看原图
1
2
1
1
2
Adorable dog celebrates a bit too soon
2
1
1
Adorable dog celebrates a bit too soon
"In a beginner obedience competition, Molly was supposed to sit/stay and wait for me to call her to me. Then she was to sit in front of me and look up. This is the last exercise of the event and Molly decided to celebrate early. We lost a…
有趣
(
twitter.com)
00:00:39
3
ELON: PHOTONS ARE ALL YOU NEED TO TRAIN THE AI
2
1
1
ELON: PHOTONS ARE ALL YOU NEED TO TRAIN THE AI
“The really wild thing about [Tesla’s] end-to-end training is that it learns to read, like it can read signs, but we never taught it to read.
We never taught it what a car was or what a person was or a cyclist.
It learned what
btc
(
twitter.com)
00:00:43
4
2
1
1
5
2
1
1
6
2
1
1
7
2
1
1
8
2
1
1
9
2
1
1
10
2
1
1
11
3
2
2
12
2
1
1
13
2
1
1
14
2
1
1
15
2
1
1
16
2
1
1
17
2
1
1
18
2
1
1
19
2
1
1
20
2
1
1
21
2
1
1
22
2
1
1
23
2
1
1
24
2
1
1