Andrej Karpathy
|
492b79fb31
|
get rid of spurious function for the model
|
2020-08-30 11:39:55 -07:00 |
|
Andrej Karpathy
|
d91bb1c0be
|
make labels non-blocking transfer to overlap them, but i don't really expect this to do too much to latency
|
2020-08-30 11:11:46 -07:00 |
|
Andrej Karpathy
|
4817231b23
|
testing now works with both lightning and minLightning
|
2020-08-30 11:11:17 -07:00 |
|
Andrej Karpathy
|
9b1e5a461f
|
delete Result structs in favor of dicts
|
2020-08-30 10:46:32 -07:00 |
|
Andrej Karpathy
|
452a5ab9a0
|
massive refactor yet again. this was all probably a pretty bad idea
|
2020-08-29 23:58:45 -07:00 |
|
Andrej Karpathy
|
1aa67ca527
|
switch to a faster version of zero_grad()
|
2020-08-29 20:50:48 -07:00 |
|
Andrej Karpathy
|
ebd40f112c
|
support fp16/32 precision in bench
|
2020-08-29 17:47:06 -07:00 |
|
Andrej Karpathy
|
fa10298a8d
|
use a standard benchmark (text8) and implement train/val/test splits
|
2020-08-29 17:30:41 -07:00 |
|
Andrej Karpathy
|
fb37e03cd1
|
refactor into a datamodule, attempt number 1
|
2020-08-29 16:38:58 -07:00 |
|
Andrej Karpathy
|
81650ae4d7
|
one more refactor, this is better because the equivalence to lightning is now much cleaner and all of lightning functionality is in one file
|
2020-08-29 15:40:21 -07:00 |
|