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

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: 0.4048
• Logps: -92.9808
• Logits: -1.5197
• Objective: 0.4085
• Dpo Loss: 0.6871
• Regularize: 0.4085
• Ranking Simple: 0.5196
• Ranking Idealized: 0.5888
• Ranking Idealized Expo: 0.5103

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-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
0.4193	0.2834	50	0.4127	-91.2123	-1.4175	0.4100	0.6922	0.4100	0.5114	0.5888	0.5103
0.4027	0.5668	100	0.4078	-91.1627	-1.4426	0.4063	0.6899	0.4063	0.5114	0.5888	0.5103
0.3645	0.8503	150	0.4059	-91.6651	-1.4648	0.4065	0.6895	0.4065	0.5093	0.5888	0.5103
0.3416	1.1337	200	0.4050	-91.3921	-1.4831	0.4071	0.6885	0.4071	0.5124	0.5888	0.5103
0.33	1.4171	250	0.4063	-92.3039	-1.4859	0.4085	0.6888	0.4085	0.5103	0.5888	0.5103
0.3193	1.7005	300	0.4041	-92.1735	-1.4928	0.4069	0.6880	0.4069	0.5134	0.5888	0.5103
0.3108	1.9839	350	0.4044	-92.4054	-1.4988	0.4070	0.6875	0.4070	0.5134	0.5888	0.5103
0.3048	2.2674	400	0.4050	-92.6592	-1.5060	0.4083	0.6881	0.4083	0.5134	0.5888	0.5103
0.2719	2.5508	450	0.4046	-92.6320	-1.5051	0.4084	0.6875	0.4084	0.5186	0.5888	0.5103
0.2722	2.8342	500	0.4044	-92.6225	-1.5137	0.4081	0.6873	0.4081	0.5176	0.5888	0.5103
0.2796	3.1176	550	0.4041	-92.7195	-1.5148	0.4077	0.6873	0.4077	0.5196	0.5888	0.5103
0.2553	3.4010	600	0.4045	-92.8725	-1.5165	0.4083	0.6872	0.4083	0.5186	0.5888	0.5103
0.252	3.6845	650	0.4046	-92.9877	-1.5188	0.4083	0.6871	0.4083	0.5196	0.5888	0.5103
0.2455	3.9679	700	0.4052	-93.0832	-1.5187	0.4088	0.6873	0.4088	0.5186	0.5888	0.5103
0.2417	4.2513	750	0.4047	-92.9650	-1.5192	0.4086	0.6872	0.4086	0.5196	0.5888	0.5103
0.2513	4.5347	800	0.4047	-92.9578	-1.5198	0.4085	0.6871	0.4085	0.5196	0.5888	0.5103
0.2539	4.8181	850	0.4048	-92.9807	-1.5197	0.4085	0.6871	0.4085	0.5196	0.5888	0.5103

Framework versions

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