qwen2.5-0.5b-expo-DPO-EXPERIMENT-1K-5e6

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: 1521.1873
• Logps: -79.1168
• Logits: -1.0707
• Objective: 1520.4889
• Dpo Loss: 1520.4889
• Regularize: 1520.4889
• Ranking Simple: 0.5258
• Ranking Idealized: 0.5093
• Ranking Idealized Expo: 0.5093

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: 6
• gradient_accumulation_steps: 12
• total_train_batch_size: 288
• total_eval_batch_size: 24
• 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	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
932.0532	0.2834	50	928.5909	-90.5890	-1.5321	972.7211	972.7211	972.7211	0.5103	0.5093	0.5093
1035.9887	0.5668	100	1589.5358	-80.1508	-1.3577	1629.7952	1629.7952	1629.7952	0.5145	0.5093	0.5093
835.8459	0.8503	150	1554.2150	-79.1304	-1.1902	1554.7245	1554.7245	1554.7245	0.5238	0.5093	0.5093
353.4232	1.1337	200	1601.4404	-77.8605	-1.1493	1618.9882	1618.9882	1618.9882	0.5279	0.5093	0.5093
363.333	1.4171	250	1571.6953	-78.8053	-1.0661	1577.6245	1577.6245	1577.6245	0.5227	0.5093	0.5093
267.0769	1.7005	300	1533.1350	-79.3922	-1.0587	1538.6410	1538.6410	1538.6410	0.5227	0.5093	0.5093
287.4463	1.9839	350	1521.1865	-79.1168	-1.0707	1520.4884	1520.4884	1520.4884	0.5258	0.5093	0.5093

Framework versions

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