qwen2.5-0.5b-expo-L2EXPO-25-8

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: 0.5029
• Objective: 0.4944
• Reward Accuracy: 0.6079
• Logp Accuracy: 0.5755
• Log Diff Policy: 9.1109
• Chosen Logps: -174.4573
• Rejected Logps: -183.5682
• Chosen Rewards: -0.8698
• Rejected Rewards: -0.9577
• Logits: -2.2881

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: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

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
0.4963	0.3154	100	0.5019	0.4943	0.5861	0.5375	2.0366	-100.3426	-102.3792	-0.1287	-0.1458	-1.3133
0.4525	0.6307	200	0.4886	0.4819	0.6085	0.5721	4.4990	-127.0285	-131.5274	-0.3955	-0.4373	-1.5453
0.4258	0.9461	300	0.4814	0.4729	0.6174	0.5895	6.1183	-135.1365	-141.2549	-0.4766	-0.5345	-1.8098
0.4056	1.2615	400	0.4869	0.4751	0.6292	0.5867	7.5129	-143.3356	-150.8486	-0.5586	-0.6305	-1.8987
0.3918	1.5769	500	0.4866	0.4788	0.6208	0.5839	7.5316	-147.2415	-154.7731	-0.5977	-0.6697	-2.0209
0.3546	1.8922	600	0.4929	0.4876	0.6107	0.5761	7.9070	-158.6364	-166.5434	-0.7116	-0.7874	-2.1730
0.301	2.2076	700	0.4937	0.4871	0.6057	0.5744	7.8447	-161.5585	-169.4032	-0.7408	-0.8160	-2.1900
0.2866	2.5230	800	0.4984	0.4930	0.6113	0.5733	8.3379	-168.4397	-176.7776	-0.8096	-0.8898	-2.2740
0.2571	2.8384	900	0.5027	0.4964	0.6119	0.5694	8.8986	-174.1609	-183.0595	-0.8669	-0.9526	-2.2442

Framework versions

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