Video wird geladen...
Video konnte nicht geladen werden
the FP8 values in your model after 50 layers of quantize/dequantize operations
102,613 Aufrufe • vor 1 Jahr •via X (Twitter)
10 Kommentare

tenderizzationvor 1 Jahr
ty @Norapom04 for finding the clip

adheepvor 1 Jahr
Wait WTF it ended right

Vivvor 1 Jahr
the hidden residual stream came through in the end

adheepvor 1 Jahr
That third kid fucked it up

gslallervor 1 Jahr
Is this an illustration of the U-Net architecture?

mobichamvor 1 Jahr
Waiting for the FP4 meme version 👀

cvor 1 Jahr
lmfao

Lost(e/employed)vor 1 Jahr
Isn’t quantization aware training adapted when they release a quantised model?

secempvor 1 Jahr
do 1bit next

Satyanshvor 1 Jahr
chinese to japanese transition was smooth in the middle lmao
