issues
search
ContextualAI
/
HALOs
A library with extensible implementations of DPO, KTO, PPO, ORPO, and other human-aware loss functions (HALOs).
https://arxiv.org/abs/2402.01306
Apache License 2.0
712
stars
40
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
FSDP hanging when training multi-GPU
#25
Phuc3010
closed
3 days ago
0
The KL value is abnormal
#24
kyzhouhzau
opened
1 week ago
1
Process oasst dataset
#23
shuoYan97
closed
3 months ago
2
added support for loading safetensor weights
#22
roshansridhar
closed
4 months ago
1
Removed outdated comments
#21
samuelzxu
closed
4 months ago
1
Hidden State mapping to two value nodes instead of 1
#20
samuelzxu
opened
4 months ago
0
Llama 3 compatibility
#19
roshansridhar
closed
4 months ago
0
Compatibility with quantized embeddings
#18
kawlil
opened
5 months ago
2
Comments in KTO Trainer `forward()`
#17
samuelzxu
closed
1 week ago
1
Request for details and assistance on PPO Experiments with SFT+PPO training
#16
roshansridhar
opened
5 months ago
1
Gradient Clipping for FSDP
#15
YJWon99
closed
6 months ago
1
ERROR:None of the inputs have requires_grad=True. Gradients will be None
#14
Pattaro
closed
6 months ago
12
Is there a problem with training?
#13
Pattaro
closed
6 months ago
4
Can you provide a clear description of the dataset structure we can use for our custom dataset.
#12
sankydesai
closed
6 months ago
2
Error fix for evaluation script `eval.py`
#11
YJWon99
closed
8 months ago
1
Eval instructions
#10
Muennighoff
opened
8 months ago
2
Losses list appears to be empty for loss=DPO
#9
abacaj
closed
8 months ago
2
How to setup the Environment without Conda?
#8
dardodel
closed
6 months ago
3
How to sample from HF models?
#7
likenneth
closed
8 months ago
3
Would you please support LoRA training?
#6
mihara-bot
closed
1 week ago
2
a few queries
#5
oapandit
closed
8 months ago
2
Question about the KL term in the loss function
#4
rajat95
closed
9 months ago
3
No checkpoint for archangel_sft-dpo_llama7b
#3
nlee-208
closed
9 months ago
2
In dataloder
#2
patronum08
closed
9 months ago
1
Figure 5 in paper does not appear to show the SFT+KTO result
#1
edbeeching
closed
9 months ago
2