qwen2.5-0.5b-expo-DPO-ES-10

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: 21.3506
• Logps: -80.2051
• Logits: -0.6148
• Objective: 20.5661
• Dpo Loss: 20.5661
• Regularize: 20.5661
• Ranking Simple: 0.5383
• Ranking Idealized: 0.5212
• Ranking Idealized Expo: 0.5212
• Wo Beta: 6.6513

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
2.0094	0.1417	50	3.1068	-90.6242	-1.4592	3.0980	3.0980	3.0980	0.5259	0.5212	0.5212	7.7179
5.9165	0.2834	100	7.1487	-82.8335	-1.4642	7.1399	7.1399	7.1399	0.5300	0.5212	0.5212	7.4498
9.9617	0.4251	150	11.8998	-83.0745	-1.3437	11.3536	11.3536	11.3536	0.5305	0.5212	0.5212	7.2609
12.4724	0.5668	200	17.0987	-79.9360	-1.3880	16.0617	16.0617	16.0617	0.5300	0.5212	0.5212	7.2290
13.2936	0.7085	250	18.5309	-77.3150	-1.3641	17.7971	17.7971	17.7971	0.5342	0.5212	0.5212	7.2078
11.5204	0.8503	300	19.4344	-76.9798	-0.9941	18.7017	18.7017	18.7017	0.5357	0.5212	0.5212	7.0136
11.3717	0.9920	350	20.3959	-76.1623	-1.0426	19.0398	19.0398	19.0398	0.5409	0.5212	0.5212	7.0261
7.0971	1.1337	400	21.9279	-76.1458	-0.6236	21.6902	21.6902	21.6902	0.5388	0.5212	0.5212	7.1227
7.5725	1.2754	450	20.9480	-76.3924	-0.8352	20.3853	20.3853	20.3853	0.5373	0.5212	0.5212	6.8500
7.6466	1.4171	500	20.9821	-80.7806	-0.7483	20.2651	20.2651	20.2651	0.5326	0.5212	0.5212	6.8824
6.9565	1.5588	550	21.3506	-80.2051	-0.6148	20.5661	20.5661	20.5661	0.5383	0.5212	0.5212	6.6513
6.7183	1.7005	600	21.1265	-78.5344	-0.6067	20.0027	20.0027	20.0027	0.5367	0.5212	0.5212	6.6768
6.9931	1.8422	650	22.2083	-77.6509	-0.5872	21.4455	21.4455	21.4455	0.5383	0.5212	0.5212	6.8190
6.1685	1.9839	700	22.3607	-77.1493	-0.5436	21.5512	21.5512	21.5512	0.5404	0.5212	0.5212	6.7299
3.4811	2.1256	750	21.8349	-78.9312	-0.7313	21.1379	21.1379	21.1379	0.5424	0.5212	0.5212	6.8213
3.3995	2.2674	800	21.3539	-79.7115	-0.5475	20.4532	20.4532	20.4532	0.5362	0.5212	0.5212	6.6867

Framework versions

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