qwen2.5-0.5b-expo-L2EXPO-EXPERIMENT-0.5-1e6

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: 0.7095
• Logps: -87.7114
• Logits: -1.2689
• Objective: 0.7117
• Dpo Loss: 0.7514
• Regularize: 0.7117
• Ranking Simple: 0.5124
• Ranking Idealized: 0.5248
• 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: 1e-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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
0.3563	0.2834	50	0.4211	-92.3339	-1.4209	0.4280	0.6873	0.4280	0.5114	0.5248	0.5093
0.4749	0.5668	100	0.4673	-91.3111	-1.3982	0.4712	0.6958	0.4712	0.5145	0.5248	0.5093
0.5468	0.8503	150	0.5596	-91.0759	-1.3204	0.5683	0.7061	0.5683	0.5145	0.5248	0.5093
0.5501	1.1337	200	0.6048	-89.7055	-1.3004	0.5995	0.7155	0.5995	0.5062	0.5248	0.5093
0.5045	1.4171	250	0.6191	-88.7256	-1.3194	0.6397	0.7337	0.6397	0.5176	0.5248	0.5093
0.5191	1.7005	300	0.6502	-87.6189	-1.3167	0.6544	0.7342	0.6544	0.5165	0.5248	0.5093
0.4473	1.9839	350	0.6869	-88.6205	-1.3002	0.6878	0.7476	0.6878	0.5093	0.5248	0.5093
0.3926	2.2674	400	0.7087	-87.7933	-1.2740	0.7147	0.7519	0.7147	0.5114	0.5248	0.5093
0.3583	2.5508	450	0.6997	-87.7180	-1.2638	0.7073	0.7468	0.7073	0.5093	0.5248	0.5093
0.2969	2.8342	500	0.7206	-87.5993	-1.2820	0.7300	0.7570	0.7300	0.5134	0.5248	0.5093
0.2456	3.1176	550	0.7082	-87.4857	-1.2747	0.7095	0.7502	0.7095	0.5134	0.5248	0.5093
0.2121	3.4010	600	0.7150	-87.8251	-1.2611	0.7195	0.7513	0.7195	0.5124	0.5248	0.5093
0.1721	3.6845	650	0.7181	-87.5542	-1.2667	0.7210	0.7529	0.7210	0.5124	0.5248	0.5093
0.1386	3.9679	700	0.7065	-87.5438	-1.2654	0.7094	0.7514	0.7094	0.5114	0.5248	0.5093
0.0985	4.2513	750	0.7096	-87.6431	-1.2699	0.7118	0.7509	0.7118	0.5145	0.5248	0.5093
0.0882	4.5347	800	0.7119	-87.7428	-1.2693	0.7145	0.7520	0.7145	0.5114	0.5248	0.5093
0.0796	4.8181	850	0.7095	-87.7155	-1.2689	0.7118	0.7515	0.7118	0.5124	0.5248	0.5093

Framework versions

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