qwen2.5-0.5b-expo-L2EXPO-EXPERIMENT-0.05-1e6

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: 0.4083
• Logps: -95.9768
• Logits: -1.6921
• Objective: 0.4121
• Dpo Loss: 0.6843
• Regularize: 0.4121
• Ranking Simple: 0.5207
• 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: 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
0.4009	0.2834	50	0.4077	-90.3481	-1.5066	0.4091	0.6906	0.4091	0.5145	0.6570	0.5114
0.3456	0.5668	100	0.4037	-92.6246	-1.6104	0.4081	0.6867	0.4081	0.5207	0.6570	0.5114
0.2786	0.8503	150	0.4061	-94.6236	-1.6473	0.4131	0.6873	0.4131	0.5207	0.6570	0.5114
0.2075	1.1337	200	0.4085	-95.7674	-1.6490	0.4120	0.6856	0.4120	0.5176	0.6570	0.5114
0.1852	1.4171	250	0.4045	-95.1014	-1.6977	0.4080	0.6845	0.4080	0.5227	0.6570	0.5114
0.172	1.7005	300	0.4055	-95.9442	-1.6403	0.4098	0.6843	0.4098	0.5227	0.6570	0.5114
0.1504	1.9839	350	0.4066	-96.3838	-1.6735	0.4094	0.6840	0.4094	0.5196	0.6570	0.5114
0.1241	2.2674	400	0.4076	-95.9834	-1.6893	0.4112	0.6844	0.4112	0.5238	0.6570	0.5114
0.1083	2.5508	450	0.4061	-96.4275	-1.6814	0.4094	0.6838	0.4094	0.5196	0.6570	0.5114
0.0989	2.8342	500	0.4076	-95.7645	-1.6797	0.4115	0.6844	0.4115	0.5176	0.6570	0.5114
0.0857	3.1176	550	0.4070	-96.7057	-1.6864	0.4108	0.6841	0.4108	0.5196	0.6570	0.5114
0.0723	3.4010	600	0.4083	-96.7714	-1.6934	0.4112	0.6840	0.4112	0.5227	0.6570	0.5114
0.0603	3.6845	650	0.4085	-95.6858	-1.6889	0.4126	0.6846	0.4126	0.5207	0.6570	0.5114
0.0658	3.9679	700	0.4086	-95.9264	-1.6962	0.4119	0.6843	0.4119	0.5217	0.6570	0.5114
0.0521	4.2513	750	0.4083	-95.9188	-1.6900	0.4119	0.6843	0.4119	0.5227	0.6570	0.5114
0.0529	4.5347	800	0.4081	-95.8100	-1.6918	0.4119	0.6843	0.4119	0.5207	0.6570	0.5114
0.0471	4.8181	850	0.4083	-95.9782	-1.6920	0.4121	0.6844	0.4121	0.5196	0.6570	0.5114

Framework versions

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