Falcon-7B-Instruct-ORPO-SALT

This model is a fine-tuned version of tiiuae/falcon-7b-instruct on the dpo_mix_en and the bct_non_cot_dpo_1000 datasets. It achieves the following results on the evaluation set:

• Loss: 1.4485
• Rewards/chosen: -0.1373
• Rewards/rejected: -0.1429
• Rewards/accuracies: 0.4809
• Rewards/margins: 0.0056
• Logps/rejected: -1.4290
• Logps/chosen: -1.3726
• Logits/rejected: -14.3178
• Logits/chosen: -14.2778
• Sft Loss: 1.3726
• Odds Ratio Loss: 0.7590

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen	Sft Loss	Odds Ratio Loss
1.5005	0.8082	500	1.5202	-0.1444	-0.1490	0.4818	0.0046	-1.4898	-1.4436	-14.2657	-14.2276	1.4436	0.7658
1.3401	1.6165	1000	1.4635	-0.1387	-0.1442	0.4836	0.0055	-1.4423	-1.3875	-14.3083	-14.2685	1.3875	0.7603
1.446	2.4247	1500	1.4485	-0.1373	-0.1429	0.4809	0.0056	-1.4290	-1.3726	-14.3178	-14.2778	1.3726	0.7590

Framework versions

• PEFT 0.10.0
• Transformers 4.40.1
• Pytorch 2.3.0
• Datasets 2.19.0
• Tokenizers 0.19.1