heiway
|
d76b9174d1
fix a dtype mismatch when use mps (#790)
|
1 سال پیش |
Whale and Dolphin
|
40665e1a39
[fix]fix problems to let version 1.5 support sft (#774)
|
1 سال پیش |
spicysama
|
b951de3b72
v1.5 (#696)
|
1 سال پیش |
PoTaTo
|
834b07257c
Agent inference (#650)
|
1 سال پیش |
spicysama
|
e37a445f51
Fix backend (#627)
|
1 سال پیش |
bfs18
|
dfc95d0f5c
Add preprocessing and validation steps when loading the checkpoint. (#433)
|
1 سال پیش |
spicysama
|
ea53678446
Quantization Support (#316)
|
1 سال پیش |
Lengyue
|
c9209e94d9
support finetuning
|
1 سال پیش |
spicysama
|
584be1d047
Adaptation to version 1.2 (#301)
|
1 سال پیش |
Leng Yue
|
5e7914472f
This PR brings V1.2 inference into main (#300)
|
1 سال پیش |
Lengyue
|
26af12b8c6
Make AR and naive decoder configurable
|
2 سال پیش |
Lengyue
|
3d002fd18b
Optimize amp training
|
2 سال پیش |
Lengyue
|
87ca8cf62e
Rollback decoding
|
2 سال پیش |
Lengyue
|
6e67ebda73
Optimize training speed by skip unused decode
|
2 سال پیش |
Lengyue
|
0a1986bb14
Fix data leaking
|
2 سال پیش |
Lengyue
|
362a1a7116
Update dual ar generate & config
|
2 سال پیش |
Lengyue
|
9e1a9debfd
fix bug & implement generate
|
2 سال پیش |
Lengyue
|
7a31b4043a
Fix target rotate
|
2 سال پیش |
Lengyue
|
0855b454e4
Add VITS vq
|
2 سال پیش |
Lengyue
|
cecee062c4
Remove flash attn deps as PyTorch ships in 2.2.0
|
2 سال پیش |
Lengyue
|
415c78a7d4
Disable neft on eval
|
2 سال پیش |
Lengyue
|
f5a2df2d23
Add 4 in 8 out support
|
2 سال پیش |
Lengyue
|
8093258065
Optimize lora & add auto dpo training
|
2 سال پیش |
Lengyue
|
7ac4d4b918
Add neft and save lora only
|
2 سال پیش |
Lengyue
|
39f6902119
Add dropout options to optimize overfitting
|
2 سال پیش |
Lengyue
|
2919eafd19
Disable flash attention 2 by default
|
2 سال پیش |
Lengyue
|
dcc5e80ce2
Add flash attention & gradient checkpointing
|
2 سال پیش |
spicysama
|
06a35aef53
add fastapi for inference (#12)
|
2 سال پیش |
Lengyue
|
c583555995
Add half precision inference and document
|
2 سال پیش |
Lengyue
|
9649873532
Update generate examples
|
2 سال پیش |