qwen2.5-0.5b-expo-L2EXPO-25-3

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft3-25-2 on the hZzy/train_pairwise_all_new4 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4834
• Objective: 0.4703
• Reward Accuracy: 0.6253
• Logp Accuracy: 0.6208
• Log Diff Policy: 104.4308
• Chosen Logps: -755.0930
• Rejected Logps: -859.5239
• Chosen Rewards: -0.6676
• Rejected Rewards: -0.7717
• Logits: -7.5473

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: 2e-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: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Objective	Reward Accuracy	Logp Accuracy	Log Diff Policy	Chosen Logps	Rejected Logps	Chosen Rewards	Rejected Rewards	Logits
0.5021	0.1577	50	0.5103	0.5036	0.5436	0.5414	2.5654	-115.5719	-118.1373	-0.0281	-0.0303	-1.5308
0.5018	0.3154	100	0.5003	0.4908	0.5872	0.5845	21.3523	-263.8485	-285.2008	-0.1764	-0.1974	-3.0190
0.504	0.4731	150	0.4872	0.4773	0.6063	0.6001	44.6167	-383.3965	-428.0132	-0.2959	-0.3402	-4.3297
0.4543	0.6307	200	0.4813	0.4703	0.6180	0.6202	61.6627	-484.1489	-545.8116	-0.3967	-0.4580	-5.3493
0.4567	0.7884	250	0.4788	0.4692	0.6247	0.6247	71.4366	-527.9006	-599.3371	-0.4404	-0.5115	-6.1216
0.4225	0.9461	300	0.4779	0.4668	0.6208	0.6163	87.7644	-624.8009	-712.5652	-0.5373	-0.6248	-6.6622
0.4	1.1038	350	0.4803	0.4703	0.6169	0.6119	88.7974	-605.5159	-694.3134	-0.5180	-0.6065	-6.4728
0.3985	1.2615	400	0.4814	0.4707	0.6197	0.6163	94.6204	-706.7501	-801.3705	-0.6193	-0.7136	-7.3144
0.3723	1.4192	450	0.4802	0.4719	0.6163	0.6147	94.9399	-637.5876	-732.5275	-0.5501	-0.6447	-6.9109
0.381	1.5769	500	0.4905	0.4779	0.6107	0.6113	105.0160	-843.2735	-948.2895	-0.7558	-0.8605	-7.7570
0.3538	1.7346	550	0.4837	0.4721	0.6130	0.6130	95.3465	-707.1511	-802.4977	-0.6197	-0.7147	-7.6110
0.347	1.8922	600	0.4822	0.4705	0.6180	0.6158	98.6297	-755.7971	-854.4269	-0.6683	-0.7666	-7.5941

Framework versions

• Transformers 4.42.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.19.1