正在加载视频...

视频加载失败

the FP8 values in your model after 50 layers of quantize/dequantize operations

102,613 次观看 • 1 年前 •via X (Twitter)

10 条评论

tenderizzation 的头像
tenderizzation1 年前

ty @Norapom04 for finding the clip

adheep 的头像
adheep1 年前

Wait WTF it ended right

Viv 的头像
Viv1 年前

the hidden residual stream came through in the end

adheep 的头像
adheep1 年前

That third kid fucked it up

gslaller 的头像
gslaller1 年前

Is this an illustration of the U-Net architecture?

mobicham 的头像
mobicham1 年前

Waiting for the FP4 meme version 👀

c 的头像
c1 年前

lmfao

Lost(e/employed) 的头像
Lost(e/employed)1 年前

Isn’t quantization aware training adapted when they release a quantised model?

secemp 的头像
secemp1 年前

do 1bit next

Satyansh 的头像
Satyansh1 年前

chinese to japanese transition was smooth in the middle lmao

相关视频