Issues
- 7
compute_logps function, why does it return also prob for the last token of answer
#35 opened by javismiles - 2
Could you please tell me which OpenAI API you used during the MT-Bench evaluation?
#34 opened by hitszxs - 1
Can you please give us a guide line to use your method to train on the LLAMA factory?
#20 opened by TonyQJH - 2
- 11
- 1
[Question] how does orpo combine dpo into sft ?
#33 opened by wj-Mcat - 16
Loss device for ORPOTrainer
#18 opened by ganeshkrishnan1 - 2
Discarding the prompt tokens only with the positive labels and not with the negative ones
#32 opened by javismiles - 2
- 2
[Question] ORPO Fine-tuning Data Format
#30 opened by nooobodynose - 4
Unexpected results using ORPO trl
#28 opened by celsowm - 3
Poor performance on llama3
#27 opened by JasonZhu1313 - 1
no reference model?
#23 opened by kxleee - 2
- 1
Memory Consumption
#22 opened by paulcx - 1
- 2
how to do ORPO with ShareGPT data?
#11 opened by pabl-o-ce - 3
attention mask in compute_logps function
#17 opened by hjc3613 - 3
how to install requirements.txt on colab?
#12 opened by srn-source - 1
[Question] Memory requiremsnts for ORPOTrainer
#13 opened by snassimr - 1
[Question] ORPO + SFTTrainer + QLora
#10 opened by snassimr - 3
- 2
- 12
prompt formatting issue
#5 opened by RonanKMcGovern - 11
- 7
4xA6000 training - failed to save model
#3 opened by rkinas