qwen2.5-0.5b-expo-L1EXPO-25-3

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-25-1 on the hZzy/train_pairwise_strong_new dataset. It achieves the following results on the evaluation set:

• Loss: 0.0449
• Objective: 0.0464
• Ranking Simple: 0.4851
• Reward Accuracy: 0.5163
• Logp Accuracy: 0.4851
• Log Diff Policy: -0.0199
• Chosen Logps: -94.0044
• Rejected Logps: -93.9845
• Chosen Rewards: 0.1334
• Rejected Rewards: 0.1295
• Logits: -1.1553

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: 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: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Objective	Ranking Simple	Reward Accuracy	Logp Accuracy	Log Diff Policy	Chosen Logps	Rejected Logps	Chosen Rewards	Rejected Rewards	Logits
0.0443	0.3212	50	0.0391	0.0396	0.4878	0.5421	0.4878	-0.0011	-94.2958	-94.2947	0.1042	0.0985	-1.1971
0.0742	0.6424	100	0.0500	0.0519	0.4918	0.5027	0.4918	-0.0428	-95.4319	-95.3890	-0.0094	-0.0110	-1.1971
0.0772	0.9636	150	0.0503	0.0518	0.4918	0.5136	0.4918	-0.0491	-95.7999	-95.7508	-0.0462	-0.0471	-1.1257
0.0949	1.2848	200	0.0694	0.0724	0.4918	0.5109	0.4918	-0.0124	-94.8558	-94.8434	0.0482	0.0436	-1.1415
0.0873	1.6060	250	0.0575	0.0589	0.4932	0.5367	0.4932	-0.0147	-93.7249	-93.7102	0.1613	0.1569	-1.1543
0.0782	1.9272	300	0.0538	0.0566	0.4891	0.5204	0.4891	-0.0493	-93.3568	-93.3074	0.1981	0.1972	-1.1650
0.0549	2.2484	350	0.0482	0.0500	0.4891	0.5082	0.4891	-0.0244	-93.5386	-93.5143	0.1800	0.1765	-1.1559
0.0487	2.5696	400	0.0464	0.0477	0.4864	0.5258	0.4864	-0.0271	-93.9310	-93.9039	0.1407	0.1376	-1.1521
0.0512	2.8908	450	0.0449	0.0464	0.4864	0.5149	0.4864	-0.0210	-94.0002	-93.9792	0.1338	0.1300	-1.1554

Framework versions

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