qwen2.5-0.5b-expo-DPO-EXPERIMENT-10-5e6

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-news-IFT on the hZzy/train_pairwise dataset. It achieves the following results on the evaluation set:

• Loss: 15.2566
• Logps: -80.3981
• Logits: -1.0046
• Objective: 15.1445
• Dpo Loss: 15.1445
• Regularize: 15.1445
• Ranking Simple: 0.5134
• Ranking Idealized: 0.5093
• Ranking Idealized Expo: 0.5093

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 12
• total_train_batch_size: 288
• total_eval_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
9.5723	0.2834	50	9.2586	-89.6862	-1.4979	9.6501	9.6501	9.6501	0.5134	0.5093	0.5093
9.8364	0.5668	100	15.5453	-79.4201	-1.3475	15.5409	15.5409	15.5409	0.5176	0.5093	0.5093
8.8451	0.8503	150	16.6626	-82.1459	-1.1122	16.5626	16.5626	16.5626	0.5145	0.5093	0.5093
3.8083	1.1337	200	16.0519	-81.6751	-1.0874	16.3240	16.3240	16.3240	0.5186	0.5093	0.5093
3.6019	1.4171	250	15.8144	-81.5609	-0.9933	15.7679	15.7679	15.7679	0.5176	0.5093	0.5093
2.1682	1.7005	300	15.3824	-80.3329	-1.0036	15.2004	15.2004	15.2004	0.5114	0.5093	0.5093
2.703	1.9839	350	15.2566	-80.3981	-1.0046	15.1445	15.1445	15.1445	0.5134	0.5093	0.5093

Framework versions

• Transformers 4.42.0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1