Oleksandr Kuvshynov
|
0d8a4cf5dc
|
rename model-specific files to llama2_
I'll try to add mistral/dbrx support, and they might need a different
logic for loading/eval/backprop
|
2024-03-28 10:14:10 -04:00 |
|
Oleksandr Kuvshynov
|
4c2996d654
|
slowllama: fix merge sharding
|
2024-03-18 13:39:55 -04:00 |
|
Oleksandr Kuvshynov
|
04ea163764
|
slowllama: move logs
|
2023-09-30 01:18:35 -04:00 |
|
Oleksandr Kuvshynov
|
636fb6c47f
|
slowllama: faster merge for 70B
|
2023-09-19 20:11:08 -04:00 |
|
Oleksandr Kuvshynov
|
04e6c37040
|
slowllama: better save
|
2023-09-18 09:34:04 -04:00 |
|
Oleksandr Kuvshynov
|
ddedf8478a
|
slowllama: fix OOM for saving 70B models
|
2023-09-16 19:28:18 -04:00 |
|
Oleksandr Kuvshynov
|
c67074291d
|
slowllama: merge lora back
|
2023-09-15 11:03:29 -04:00 |