qwen2.5-0.5b-expo-DPO-L2EXPO-W2-noES-0.1

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-news-IFT on the hZzy/train_pairwise_weighted dataset. It achieves the following results on the evaluation set:

• Loss: 0.1180
• Logps: -80.3490
• Logits: -0.5109
• Objective: 0.1149
• Dpo Loss: 0.7081
• Regularize: 0.6066
• Ranking Simple: 0.5362

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: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple
0.0974	0.1417	50	0.1013	-88.6215	-1.4569	0.1014	0.6831	0.4279	0.5285
0.0968	0.2834	100	0.1060	-85.6897	-1.5103	0.1055	0.6828	0.4821	0.5300
0.0947	0.4251	150	0.1114	-83.2094	-1.5092	0.1107	0.6962	0.5376	0.5326
0.0964	0.5668	200	0.1115	-77.7950	-1.1358	0.1114	0.6928	0.5439	0.5316
0.0994	0.7085	250	0.1146	-78.1513	-1.0086	0.1126	0.7003	0.5531	0.5373
0.1059	0.8503	300	0.1161	-81.8632	-0.8100	0.1133	0.7048	0.5859	0.5450
0.0992	0.9920	350	0.1190	-79.1597	-0.8348	0.1167	0.7056	0.6018	0.5461
0.0992	1.1337	400	0.1205	-80.4108	-0.6239	0.1171	0.7107	0.6249	0.5409
0.0817	1.2754	450	0.1204	-80.6768	-0.6133	0.1169	0.7151	0.6378	0.5378
0.097	1.4171	500	0.1191	-79.8564	-0.6515	0.1164	0.7078	0.6180	0.5383
0.0865	1.5588	550	0.1196	-81.0992	-0.5267	0.1169	0.7048	0.6207	0.5367
0.0763	1.7005	600	0.1193	-81.0751	-0.4880	0.1161	0.7045	0.6161	0.5399
0.0772	1.8422	650	0.1186	-80.2917	-0.4325	0.1154	0.7078	0.6067	0.5367
0.0706	1.9839	700	0.1187	-79.8271	-0.5150	0.1151	0.7099	0.6083	0.5383
0.0684	2.1256	750	0.1184	-80.5600	-0.4615	0.1149	0.7091	0.6088	0.5383
0.0655	2.2674	800	0.1189	-80.4247	-0.4822	0.1158	0.7088	0.6121	0.5383
0.0722	2.4091	850	0.1181	-80.5667	-0.4638	0.1149	0.7078	0.6067	0.5388
0.0787	2.5508	900	0.1179	-80.4408	-0.5063	0.1148	0.7076	0.6056	0.5373
0.0657	2.6925	950	0.1180	-80.4005	-0.5067	0.1149	0.7081	0.6065	0.5373
0.0583	2.8342	1000	0.1180	-80.3490	-0.5108	0.1149	0.7081	0.6066	0.5367
0.0679	2.9759	1050	0.1180	-80.3490	-0.5109	0.1149	0.7081	0.6066	0.5362

Framework versions

• Transformers 4.42.0
• Pytorch 2.3.0+cu121
• Datasets 3.2.0
• Tokenizers 0.19.1