qwen2.5-0.5b-expo-DPO-ES-1

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: 2.3420
• Logps: -83.4105
• Logits: -0.6597
• Objective: 2.2592
• Dpo Loss: 2.2592
• Regularize: 2.2592
• Ranking Simple: 0.5404
• Ranking Idealized: 0.5295
• Ranking Idealized Expo: 0.5212
• Wo Beta: 6.6836

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo	Wo Beta
0.7017	0.1417	50	0.8470	-93.0245	-1.4582	0.8570	0.8570	0.8570	0.5238	0.5295	0.5212	7.8506
0.8112	0.2834	100	1.0529	-86.6804	-1.4383	1.0274	1.0274	1.0274	0.5285	0.5295	0.5212	7.4982
1.0895	0.4251	150	1.4498	-84.4335	-1.2965	1.4010	1.4010	1.4010	0.5321	0.5295	0.5212	7.2692
1.2362	0.5668	200	1.7035	-77.7194	-1.2955	1.6116	1.6116	1.6116	0.5321	0.5295	0.5212	7.2264
1.3151	0.7085	250	1.9222	-92.7224	-1.2565	1.8319	1.8319	1.8319	0.5311	0.5295	0.5212	7.1855
1.1899	0.8503	300	2.0297	-90.9351	-0.9786	1.9587	1.9587	1.9587	0.5367	0.5295	0.5212	6.9337
1.1441	0.9920	350	2.1653	-82.1291	-1.0211	2.0545	2.0545	2.0545	0.5424	0.5295	0.5212	7.0017
0.725	1.1337	400	2.2886	-84.3458	-0.7529	2.2360	2.2360	2.2360	0.5331	0.5295	0.5212	7.1541
0.7626	1.2754	450	2.1595	-80.5955	-0.8863	2.0657	2.0657	2.0657	0.5326	0.5295	0.5212	6.7939
0.8048	1.4171	500	2.2134	-82.3489	-0.7432	2.0975	2.0975	2.0975	0.5342	0.5295	0.5212	6.7984
0.7106	1.5588	550	2.1705	-85.0673	-0.6665	2.0696	2.0696	2.0696	0.5321	0.5295	0.5212	6.8614
0.6934	1.7005	600	2.2127	-81.6773	-0.7358	2.0693	2.0693	2.0693	0.5362	0.5295	0.5212	6.7265
0.6885	1.8422	650	2.2198	-82.8870	-0.6787	2.1432	2.1432	2.1432	0.5362	0.5295	0.5212	6.8202
0.6477	1.9839	700	2.3420	-83.4105	-0.6597	2.2592	2.2592	2.2592	0.5404	0.5295	0.5212	6.6836
0.3785	2.1256	750	2.2919	-84.0369	-0.7841	2.2005	2.2005	2.2005	0.5435	0.5295	0.5212	6.8514
0.3316	2.2674	800	2.2220	-84.2990	-0.6767	2.1123	2.1123	2.1123	0.5409	0.5295	0.5212	6.7663
0.3283	2.4091	850	2.3020	-85.0834	-0.6538	2.2212	2.2212	2.2212	0.5409	0.5295	0.5212	6.7773
0.3516	2.5508	900	2.2723	-84.7564	-0.6225	2.1911	2.1911	2.1911	0.5362	0.5295	0.5212	6.8162
0.3245	2.6925	950	2.3304	-83.6421	-0.7129	2.2523	2.2523	2.2523	0.5336	0.5295	0.5212	6.8942

Framework versions

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