qwen2.5-0.5b-expo-DPO-L2EXPO-W0-noES2-0.1

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: 324.6227
• Logps: -88.5336
• Logits: -1.1616
• Objective: 321.3043
• Dpo Loss: 0.6772
• Ranking Simple: 0.5471

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
298.227	0.1417	50	317.0191	-89.8084	-1.5244	315.7606	0.6789	0.5311
275.3632	0.2834	100	315.6042	-94.9782	-1.5619	311.9210	0.6685	0.5430
263.5541	0.4251	150	323.5653	-86.5190	-1.4200	317.0345	0.6707	0.5466
251.2096	0.5668	200	331.8273	-95.4989	-1.6021	324.8527	0.6841	0.5455
260.8009	0.7085	250	335.3750	-89.0524	-1.3996	327.2677	0.6886	0.5383
241.17	0.8503	300	326.9033	-86.8043	-1.3637	321.4521	0.6751	0.5518
217.6821	0.9920	350	330.4503	-84.6762	-1.1931	320.4708	0.6764	0.5497
196.3838	1.1337	400	331.8378	-85.2363	-1.1204	324.6776	0.6834	0.5492
196.8245	1.2754	450	328.4219	-87.7967	-1.1766	321.0422	0.6758	0.5502
206.0427	1.4171	500	327.5086	-86.1759	-1.1174	321.6342	0.6782	0.5502
185.0637	1.5588	550	325.0724	-92.8798	-1.1304	320.1057	0.6694	0.5554
183.7322	1.7005	600	324.3962	-89.2066	-0.9798	319.9032	0.6742	0.5554
206.6587	1.8422	650	324.2286	-88.9897	-1.1055	319.4041	0.6736	0.5533
188.4019	1.9839	700	323.3723	-89.3024	-1.0999	318.8545	0.6727	0.5497
165.7588	2.1256	750	324.4387	-89.4604	-1.1334	320.7416	0.6772	0.5502
164.7524	2.2674	800	323.6566	-89.3705	-1.1091	320.3278	0.6756	0.5497
160.4428	2.4091	850	324.1759	-88.9819	-1.1590	321.1419	0.6765	0.5461
164.2802	2.5508	900	324.5910	-88.9856	-1.1737	321.6709	0.6779	0.5461
168.6074	2.6925	950	324.7719	-88.5500	-1.1605	321.5456	0.6777	0.5461
165.6921	2.8342	1000	324.6280	-88.5341	-1.1614	321.3309	0.6773	0.5461
160.429	2.9759	1050	324.6226	-88.5336	-1.1616	321.3043	0.6772	0.5471

Framework versions

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