|
ckpt_convert
|
1. llama_model.py attention实现回退
|
2023-07-06 10:33:21 +08:00 |
|
eval_harness
|
create megatron core
|
2023-07-24 15:00:57 +08:00 |
|
glue
|
change megatron to ascendspeed
|
2023-06-10 21:26:01 +08:00 |
|
orqa
|
change megatron to ascendspeed
|
2023-06-10 21:26:01 +08:00 |
|
race
|
change megatron to ascendspeed
|
2023-06-10 21:26:01 +08:00 |
|
vision
|
create megatron core
|
2023-07-24 15:00:57 +08:00 |
|
zeroshot
|
create megatron core
|
2023-07-24 15:00:57 +08:00 |
|
data_utils.py
|
fork megatron-deepspeed code.
|
2023-05-25 14:49:59 +08:00 |
|
ensemble_classifier.py
|
fork megatron-deepspeed code.
|
2023-05-25 14:49:59 +08:00 |
|
eval_utils.py
|
create megatron core
|
2023-07-24 15:00:57 +08:00 |
|
finetune_utils.py
|
create megatron core
|
2023-07-24 15:00:57 +08:00 |
|
main.py
|
llama-zeroshot任务精度适配，对齐源论文中的效果。
|
2023-06-25 09:34:42 +08:00 |
