qwen2.5-0.5b-expo-DPO-noES5-1

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: 1.9399
• Logps: -81.1684
• Logits: -0.8509
• Objective: 1.8787
• Dpo Loss: 1.8787
• Ranking Simple: 0.5347

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: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Ranking Simple
1.1219	0.1417	50	1.1407	-91.1504	-1.3894	1.1280	1.1280	0.5274
1.3821	0.2834	100	1.5304	-81.2234	-1.3774	1.4947	1.4947	0.5290
1.4062	0.4251	150	1.8818	-79.7787	-1.1641	1.8192	1.8192	0.5430
1.2275	0.5668	200	2.0358	-77.9854	-1.1289	1.9717	1.9717	0.5347
1.1914	0.7085	250	2.0084	-78.3385	-1.0883	1.9461	1.9461	0.5347
1.0378	0.8503	300	2.0918	-83.4707	-0.9324	2.0357	2.0357	0.5352
0.8334	0.9920	350	2.1143	-81.1740	-0.8755	1.9975	1.9975	0.5388
0.4251	1.1337	400	2.0641	-81.1689	-0.8003	2.0241	2.0241	0.5435
0.3886	1.2754	450	2.0085	-79.8813	-0.8999	1.9598	1.9598	0.5388
0.4352	1.4171	500	2.0449	-80.7357	-0.8634	1.9819	1.9819	0.5367
0.3103	1.5588	550	1.9784	-80.8827	-0.8672	1.9073	1.9073	0.5373
0.2489	1.7005	600	1.9488	-81.0833	-0.8421	1.8851	1.8851	0.5367
0.3631	1.8422	650	1.9417	-81.1721	-0.8529	1.8805	1.8805	0.5347
0.3009	1.9839	700	1.9399	-81.1684	-0.8509	1.8787	1.8787	0.5347

Framework versions

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