qwen2.5-0.5b-expo-L2EXPO-W0-noES5-0.05

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: 179.0654
• Logps: -93.4070
• Logits: -1.6430
• Objective: 176.1436
• Dpo Loss: 0.6785
• Regularize: 0.3996
• Ranking Simple: 0.5342
• Wo Beta: 17.0972

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: 1e-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	Wo Beta
181.8716	0.1417	50	182.0715	-90.3592	-1.4456	180.1049	0.6894	0.4083	0.5269	16.4217
157.5861	0.2834	100	180.9991	-91.7146	-1.5597	179.7807	0.6852	0.4085	0.5388	16.6421
147.4354	0.4251	150	179.6252	-88.3708	-1.5794	177.2645	0.6811	0.4007	0.5383	17.0744
138.6462	0.5668	200	179.4377	-90.4436	-1.5522	175.5756	0.6805	0.3987	0.5342	16.6929
122.5651	0.7085	250	178.9847	-93.0301	-1.6388	176.6755	0.6795	0.4017	0.5300	17.1382
110.1558	0.8503	300	178.2535	-94.1108	-1.6237	176.3972	0.6795	0.4012	0.5311	17.0693
99.307	0.9920	350	180.1735	-96.0474	-1.5854	177.3710	0.6788	0.4026	0.5336	17.1627
87.6253	1.1337	400	178.9584	-94.0894	-1.5779	176.3463	0.6783	0.3997	0.5336	17.1200
77.9665	1.2754	450	178.7482	-93.8793	-1.6468	175.9861	0.6798	0.4001	0.5321	16.9068
70.9202	1.4171	500	178.7919	-94.5267	-1.6244	175.6359	0.6787	0.3987	0.5342	16.9906
68.4	1.5588	550	179.6713	-93.1219	-1.6340	175.8635	0.6783	0.3992	0.5336	17.0272
62.6522	1.7005	600	179.6970	-93.5471	-1.6273	176.5027	0.6786	0.4002	0.5362	17.1271
62.1281	1.8422	650	178.4731	-92.9689	-1.6053	175.4037	0.6782	0.3979	0.5362	17.0350
58.8228	1.9839	700	178.7972	-93.4236	-1.6276	176.0926	0.6786	0.3992	0.5362	17.0199
45.2464	2.1256	750	178.6497	-93.8837	-1.6225	175.6834	0.6780	0.3985	0.5342	16.9906
46.081	2.2674	800	179.1421	-93.3818	-1.6332	176.3395	0.6787	0.4002	0.5331	17.0616
38.6939	2.4091	850	178.9382	-93.4067	-1.6352	175.9686	0.6784	0.3992	0.5362	17.0858
39.6509	2.5508	900	179.1196	-93.3231	-1.6445	176.1842	0.6785	0.3996	0.5357	17.1093
37.5296	2.6925	950	179.0316	-93.3121	-1.6429	176.0845	0.6784	0.3994	0.5347	17.0760
38.0286	2.8342	1000	179.0622	-93.4092	-1.6428	176.1425	0.6785	0.3996	0.5342	17.0993
41.7608	2.9759	1050	179.0654	-93.4070	-1.6430	176.1437	0.6785	0.3996	0.5342	17.0972

Framework versions

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