Temp-L1-SFT-L2-KTO

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: 0.2213
• Rewards/chosen: 0.2579
• Rewards/rejected: -6.0725
• Rewards/accuracies: 1.0
• Rewards/margins: 6.3304
• Logps/rejected: -652.1185
• Logps/chosen: -0.1197
• Logits/rejected: -2.6590
• Logits/chosen: -2.5711

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
0.2255	0.2497	1000	0.2230	0.2551	-5.4032	1.0	5.6583	-585.1871	-0.3988	-2.6372	-2.5514
0.2252	0.4994	2000	0.2215	0.2576	-5.9860	1.0	6.2436	-643.4705	-0.1526	-2.6560	-2.5690
0.2264	0.7492	3000	0.2213	0.2579	-6.0565	1.0	6.3144	-650.5204	-0.1267	-2.6590	-2.5715
0.2262	0.9989	4000	0.2213	0.2579	-6.0725	1.0	6.3304	-652.1185	-0.1197	-2.6590	-2.5711

Framework versions

• PEFT 0.7.1
• Transformers 4.40.2
• Pytorch 2.1.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1