qwen_uncCPO_entropy_0_01

This model is a fine-tuned version of trl-lib/qwen1.5-0.5b-sft on the yakazimir/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.0500
• Sft Loss: 3.9220
• Rewards/chosen: -4.3252
• Rewards/rejected: -5.1044
• Rewards/accuracies: 0.6892
• Rewards/margins: 0.7793
• Logps/rejected: -5.1044
• Logps/chosen: -4.3252
• Logits/rejected: 0.1444
• Logits/chosen: 0.0509

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: 2
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Sft Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0563	0.2141	400	0.0573	4.8352	-5.7454	-6.0246	0.5445	0.2792	-6.0246	-5.7454	0.6512	0.5372
0.0533	0.4282	800	0.0524	4.2340	-4.6954	-5.0777	0.6157	0.3823	-5.0777	-4.6954	0.2939	0.1644
0.0533	0.6422	1200	0.0518	4.1504	-4.5198	-5.0186	0.6484	0.4989	-5.0186	-4.5198	0.4014	0.2684
0.0508	0.8563	1600	0.0512	4.0690	-4.5220	-5.0081	0.6491	0.4862	-5.0081	-4.5220	0.2498	0.1344
0.0529	1.0704	2000	0.0508	3.9195	-4.3917	-4.9646	0.6521	0.5729	-4.9646	-4.3917	0.3268	0.2181
0.0522	1.2845	2400	0.0504	4.1797	-4.6133	-5.2771	0.6647	0.6638	-5.2771	-4.6133	0.2727	0.1622
0.0515	1.4986	2800	0.0504	4.0933	-4.4442	-5.0786	0.6825	0.6344	-5.0786	-4.4442	0.2050	0.0984
0.0526	1.7127	3200	0.0503	4.0886	-4.4943	-5.1537	0.6751	0.6594	-5.1537	-4.4943	0.2002	0.0920
0.0533	1.9267	3600	0.0501	3.9857	-4.3809	-5.1003	0.6825	0.7195	-5.1003	-4.3809	0.1348	0.0421
0.0493	2.1408	4000	0.0500	3.9751	-4.3954	-5.1537	0.6840	0.7583	-5.1537	-4.3954	0.3029	0.1980
0.0522	2.3549	4400	0.0500	3.9820	-4.4013	-5.1632	0.6869	0.7619	-5.1632	-4.4013	0.2139	0.1131
0.0513	2.5690	4800	0.0500	3.9732	-4.3709	-5.1160	0.6944	0.7451	-5.1160	-4.3709	0.1787	0.0785
0.0498	2.7831	5200	0.0500	3.9372	-4.3318	-5.0969	0.6892	0.7651	-5.0969	-4.3318	0.2138	0.1134
0.0496	2.9972	5600	0.0500	3.9220	-4.3252	-5.1044	0.6892	0.7793	-5.1044	-4.3252	0.1444	0.0509

Framework versions

• Transformers 4.44.2
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1