qwen2.5-0.5b-expo-DPO-L2EXPO-W2-noES4-1

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: 0.6798
• Logps: -81.3660
• Logits: -0.2689
• Objective: 0.6594
• Dpo Loss: 2.5881
• Ranking Simple: 0.5290

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	Ranking Simple
0.2167	0.1417	50	0.1866	-89.7452	-1.3915	0.1841	0.8977	0.5243
0.3971	0.2834	100	0.3324	-90.0353	-1.2435	0.3345	1.3259	0.5207
0.5144	0.4251	150	0.5155	-78.4187	-1.3789	0.5175	1.9820	0.5295
0.5194	0.5668	200	0.5709	-73.3419	-0.7470	0.5515	2.1252	0.5285
0.5608	0.7085	250	0.6097	-78.0422	-0.6609	0.5942	2.2497	0.5362
0.5689	0.8503	300	0.6153	-80.6416	-0.5345	0.6046	2.2745	0.5305
0.5556	0.9920	350	0.6604	-80.0774	-0.4720	0.6456	2.4892	0.5347
0.6911	1.1337	400	0.6987	-80.2664	-0.3889	0.6688	2.6312	0.5362
0.4109	1.2754	450	0.6748	-80.4623	-0.3379	0.6533	2.6157	0.5321
0.5646	1.4171	500	0.6765	-80.2331	-0.3752	0.6533	2.5394	0.5347
0.5393	1.5588	550	0.6834	-80.9551	-0.3117	0.6649	2.5727	0.5336
0.4349	1.7005	600	0.6761	-81.5591	-0.2322	0.6615	2.5397	0.5300
0.4677	1.8422	650	0.6817	-80.5548	-0.2289	0.6641	2.5860	0.5243
0.3266	1.9839	700	0.6802	-80.5748	-0.2811	0.6562	2.5943	0.5254
0.2719	2.1256	750	0.6823	-81.2181	-0.2618	0.6627	2.5971	0.5316
0.2695	2.2674	800	0.6833	-81.1710	-0.2659	0.6653	2.5873	0.5295
0.3121	2.4091	850	0.6796	-81.3012	-0.2552	0.6616	2.5816	0.5264
0.3154	2.5508	900	0.6804	-81.3982	-0.2835	0.6608	2.5860	0.5285
0.2493	2.6925	950	0.6800	-81.4311	-0.2696	0.6596	2.5904	0.5285
0.2171	2.8342	1000	0.6800	-81.3825	-0.2689	0.6595	2.5888	0.5290
0.2548	2.9759	1050	0.6798	-81.3660	-0.2689	0.6594	2.5881	0.5290

Framework versions

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