hZzy/qwen2.5-0.5b-expo-DPO-ES-TRY0

qwen2.5-0.5b-expo-DPO-ES-TRY

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: 0.6866
• Logps: -91.4116
• Logits: -1.5339
• Objective: 0.6926
• Dpo Loss: 0.6926
• Regularize: 0.6926
• Ranking Simple: 0.5052
• Ranking Idealized: 0.5888
• Ranking Idealized Expo: 0.5093
• Dpo Wo Beta: -0.9551

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-07
• 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	Dpo Wo Beta
0.672	0.2041	36	0.6866	-91.4116	-1.5339	0.6926	0.6926	0.6926	0.5052	0.5888	0.5093	-0.9551
0.6533	0.4081	72	0.6769	-92.4758	-1.6311	0.6885	0.6885	0.6885	0.5176	0.5888	0.5093	-1.1473
0.604	0.6122	108	0.6773	-94.0584	-1.7691	0.6853	0.6853	0.6853	0.5186	0.5888	0.5093	-1.3961
0.5911	0.8162	144	0.6775	-95.7578	-1.8165	0.6886	0.6886	0.6886	0.5186	0.5888	0.5093	-1.5153

Framework versions

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