loss

300 posts

loss

loss

@dotbackward

Katılım Mayıs 2026
184 Takip Edilen18 Takipçiler
Sabitlenmiş Tweet
loss
loss@dotbackward·
GIF
ZXX
0
0
1
395
loss
loss@dotbackward·
gn
loss tweet media
0
0
0
9
loss
loss@dotbackward·
loss tweet media
ZXX
1
0
0
11
loss
loss@dotbackward·
loss tweet media
ZXX
0
0
0
7
loss
loss@dotbackward·
ima posting archives
English
0
0
0
4
loss
loss@dotbackward·
stemzz 🖤🖤
loss tweet media
Türkçe
0
0
0
5
loss
loss@dotbackward·
SEEYA #fucker
loss tweet media
हिन्दी
0
0
0
21
loss
loss@dotbackward·
so #qwen u mask all tokens except the response and im end token right?? this shit so confusing. everymodel is different
English
0
0
0
26
loss
loss@dotbackward·
ZXX
0
0
0
19
loss retweetledi
Bigsouptaco
Bigsouptaco@bigsouptaco·
Bigsouptaco tweet media
ZXX
1
69
868
8.4K
loss
loss@dotbackward·
oh found out that training left or right doesnt matter too much but personally right side is easier implementation wise
English
0
0
0
9
loss
loss@dotbackward·
goddam it since when did qwen train with right side padding
English
1
0
0
28