weeye (@wyetwt) 's Twitter Profile
weeye

@wyetwt

flipping qubits, overfitting models and mainly shitposting

ID: 1354376865285169160

linkhttp://weeye.vercel.app/posts calendar_today27-01-2021 10:33:10

944 Tweet

65 Takipçi

135 Takip Edilen

Chinmay Kak (@chinmaykak) 's Twitter Profile Photo

Introducing nanosft, a clean single file implementation of finetuning for chat style model. Loads gpt2-124M weights on nanogpt and does supervised finetuning using just pytorch. a side project that I made recently for some prep. link below :) qts/rts appericiated

Introducing nanosft, a clean single file implementation of finetuning for chat style model. Loads gpt2-124M weights on nanogpt and does supervised finetuning using just pytorch. 
a side project that I made recently for some prep. link below :) 
qts/rts appericiated
weeye (@wyetwt) 's Twitter Profile Photo

using hyperspheres and quaternion geodesics is so pretty here while i have no clue whether work like this has been done before but it is such a nice way of thinking!

using hyperspheres and quaternion geodesics is so pretty here 

while i have no clue whether work like this has been done before but it is such a nice way of thinking!
weeye (@wyetwt) 's Twitter Profile Photo

mathematicians used to duel in the 17th century when they didnt agree with someone (even gracefully die), now they just open this app

mathematicians used to duel in the 17th century when they didnt agree with someone (even gracefully die), now they just open this app
weeye (@wyetwt) 's Twitter Profile Photo

its pretty neat how this little thing reduces the dimensionality by such a high margin leaving mostly just the parts that you need

its pretty neat how this little thing reduces the dimensionality by such a high margin leaving mostly just the parts that you need
weeye (@wyetwt) 's Twitter Profile Photo

i always feel that i developed a special feeling while reading them initially, just until i find something cooler and it replaces the older one