 Lengyue
|
26af12b8c6
Make AR and naive decoder configurable
|
il y a 2 ans |
|
Lengyue
|
3d002fd18b
Optimize amp training
|
il y a 2 ans |
|
Lengyue
|
87ca8cf62e
Rollback decoding
|
il y a 2 ans |
|
Lengyue
|
6e67ebda73
Optimize training speed by skip unused decode
|
il y a 2 ans |
|
Lengyue
|
0a1986bb14
Fix data leaking
|
il y a 2 ans |
|
Lengyue
|
362a1a7116
Update dual ar generate & config
|
il y a 2 ans |
|
Lengyue
|
9e1a9debfd
fix bug & implement generate
|
il y a 2 ans |
|
Lengyue
|
7a31b4043a
Fix target rotate
|
il y a 2 ans |
|
Lengyue
|
0855b454e4
Add VITS vq
|
il y a 2 ans |
|
Lengyue
|
cecee062c4
Remove flash attn deps as PyTorch ships in 2.2.0
|
il y a 2 ans |
|
Lengyue
|
415c78a7d4
Disable neft on eval
|
il y a 2 ans |
|
Lengyue
|
f5a2df2d23
Add 4 in 8 out support
|
il y a 2 ans |
|
Lengyue
|
8093258065
Optimize lora & add auto dpo training
|
il y a 2 ans |
|
Lengyue
|
7ac4d4b918
Add neft and save lora only
|
il y a 2 ans |
|
Lengyue
|
39f6902119
Add dropout options to optimize overfitting
|
il y a 2 ans |
|
Lengyue
|
2919eafd19
Disable flash attention 2 by default
|
il y a 2 ans |
|
Lengyue
|
dcc5e80ce2
Add flash attention & gradient checkpointing
|
il y a 2 ans |
 spicysama
|
06a35aef53
add fastapi for inference (#12)
|
il y a 2 ans |
|
Lengyue
|
c583555995
Add half precision inference and document
|
il y a 2 ans |
|
Lengyue
|
9649873532
Update generate examples
|
il y a 2 ans |
|
Lengyue
|
9ac8edef1e
Support mix codebook training
|
il y a 2 ans |
|
Lengyue
|
62710e34a4
Add llama inference tool chain
|
il y a 2 ans |
|
Lengyue
|
4b22991668
Implement parallel decoding llama
|
il y a 2 ans |
|
Lengyue
|
25fce58cf6
Add llama implementation
|
il y a 2 ans |
