qwen2.5-0.5b-expo-L2EXPO-EXPERIMENT-0.05-5e7

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: 0.4176
• Logps: -99.9001
• Logits: -1.7324
• Objective: 0.4184
• Dpo Loss: 0.6853
• Regularize: 0.4184
• Ranking Simple: 0.5238
• Ranking Idealized: 0.6570
• Ranking Idealized Expo: 0.5114

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-07
• 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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
0.4122	0.2834	50	0.4102	-90.5614	-1.4615	0.4093	0.6915	0.4093	0.5124	0.6570	0.5114
0.374	0.5668	100	0.4073	-92.0361	-1.5521	0.4082	0.6883	0.4082	0.5145	0.6570	0.5114
0.3231	0.8503	150	0.4029	-92.8073	-1.6074	0.4087	0.6881	0.4087	0.5186	0.6570	0.5114
0.267	1.1337	200	0.4069	-94.7992	-1.6424	0.4110	0.6866	0.4110	0.5186	0.6570	0.5114
0.2432	1.4171	250	0.4108	-96.1389	-1.6721	0.4137	0.6877	0.4137	0.5196	0.6570	0.5114
0.2252	1.7005	300	0.4101	-95.6244	-1.6648	0.4138	0.6866	0.4138	0.5217	0.6570	0.5114
0.2082	1.9839	350	0.4102	-97.5255	-1.6989	0.4132	0.6863	0.4132	0.5196	0.6570	0.5114
0.1825	2.2674	400	0.4124	-97.7996	-1.6932	0.4144	0.6863	0.4144	0.5207	0.6570	0.5114
0.1504	2.5508	450	0.4149	-99.2029	-1.7113	0.4176	0.6864	0.4176	0.5217	0.6570	0.5114
0.1494	2.8342	500	0.4153	-99.1755	-1.7175	0.4182	0.6862	0.4182	0.5227	0.6570	0.5114
0.1407	3.1176	550	0.4161	-99.2997	-1.7183	0.4174	0.6856	0.4174	0.5217	0.6570	0.5114
0.1149	3.4010	600	0.4171	-99.9246	-1.7181	0.4181	0.6852	0.4181	0.5248	0.6570	0.5114
0.1108	3.6845	650	0.4178	-99.9118	-1.7315	0.4188	0.6853	0.4188	0.5248	0.6570	0.5114
0.1146	3.9679	700	0.4176	-99.8982	-1.7319	0.4187	0.6854	0.4187	0.5238	0.6570	0.5114
0.0986	4.2513	750	0.4175	-99.8694	-1.7322	0.4183	0.6853	0.4183	0.5238	0.6570	0.5114
0.1042	4.5347	800	0.4175	-99.8600	-1.7317	0.4183	0.6853	0.4183	0.5238	0.6570	0.5114
0.103	4.8181	850	0.4176	-99.8972	-1.7324	0.4184	0.6853	0.4184	0.5238	0.6570	0.5114

Framework versions

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