Temp-L1-SFT-L2-DPO

This model is a fine-tuned version of EllieS/TempReason-L1 on the EllieS/Temp-L2-DPO dataset. It achieves the following results on the evaluation set:

• Loss: 0.0028
• Rewards/chosen: -0.3437
• Rewards/rejected: -7.0807
• Rewards/accuracies: 1.0
• Rewards/margins: 6.7370
• Logps/rejected: -752.9371
• Logps/chosen: -60.2789
• Logits/rejected: -2.7330
• Logits/chosen: -2.6142

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
• distributed_type: multi-GPU
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0046	0.25	1000	0.0116	-0.7603	-5.8063	1.0	5.0459	-625.4973	-101.9450	-2.6955	-2.5854
0.0037	0.5	2000	0.0037	-0.2218	-6.4841	1.0	6.2624	-693.2846	-48.0898	-2.7333	-2.6137
0.0039	0.75	3000	0.0028	-0.3670	-7.1056	1.0	6.7386	-755.4296	-62.6133	-2.7342	-2.6152
0.0019	1.0	4000	0.0028	-0.3437	-7.0807	1.0	6.7370	-752.9371	-60.2789	-2.7330	-2.6142

Framework versions

• PEFT 0.7.1
• Transformers 4.39.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2