qwen2.5-0.5b-expo-L2EXPO-ES-0.01

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: 0.4116
• Logps: -92.3444
• Logits: -1.6667
• Objective: 0.4097
• Dpo Loss: 0.6906
• Regularize: 0.4097
• Ranking Simple: 0.5290
• Ranking Idealized: 0.8732
• Ranking Idealized Expo: 0.5321
• Wo Beta: 17.6306

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: 3
• gradient_accumulation_steps: 12
• total_train_batch_size: 144
• total_eval_batch_size: 12
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo	Wo Beta
0.4144	0.1417	50	0.4116	-92.3444	-1.6667	0.4097	0.6906	0.4097	0.5290	0.8732	0.5321	17.6306
0.394	0.2834	100	0.4078	-115.8757	-2.0312	0.4074	0.6863	0.4074	0.5399	0.8732	0.5321	22.1058
0.3504	0.4251	150	0.4044	-123.1670	-2.0505	0.4018	0.6803	0.4018	0.5719	0.8732	0.5321	23.5660
0.3135	0.5668	200	0.4006	-121.6031	-2.1409	0.3974	0.6781	0.3974	0.5621	0.8732	0.5321	23.0977
0.2807	0.7085	250	0.4043	-122.1639	-2.3711	0.4010	0.6790	0.4010	0.5600	0.8732	0.5321	23.6688
0.2532	0.8503	300	0.3975	-123.3866	-2.2621	0.3953	0.6769	0.3953	0.5740	0.8732	0.5321	23.6726

Framework versions

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