qwen2.5-0.5b-expo-L2EXPO-25-9

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft3-25-2 on the hZzy/direction_right2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5000
• Objective: 0.5027
• Reward Accuracy: 0.5917
• Logp Accuracy: 0.5772
• Log Diff Policy: 9.0989
• Chosen Logps: -164.6602
• Rejected Logps: -173.7591
• Chosen Rewards: -0.7687
• Rejected Rewards: -0.8563
• Logits: -2.5935

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: 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
0.5613	0.4044	100	0.5042	0.4986	0.5699	0.5721	4.3561	-116.5406	-120.8967	-0.2875	-0.3277	-1.6238
0.51	0.8089	200	0.4951	0.4916	0.5928	0.5755	7.6656	-148.1296	-155.7952	-0.6034	-0.6767	-2.0415
0.4677	1.2133	300	0.4965	0.4962	0.5956	0.5822	8.3074	-156.7594	-165.0668	-0.6897	-0.7694	-2.3321
0.4308	1.6178	400	0.4994	0.4989	0.6051	0.5805	8.7655	-165.7063	-174.4718	-0.7792	-0.8635	-2.5086

Framework versions

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