qwen_cUNL_entropy

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: 0.5196
• Rewards/chosen: -7.4572
• Rewards/rejected: -8.6117
• Rewards/accuracies: 0.7285
• Rewards/margins: 1.1545
• Logps/rejected: -8.6117
• Logps/chosen: -7.4572
• Logits/rejected: 0.5435
• Logits/chosen: 0.4914

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	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.8222	0.2141	400	0.8285	-1.7118	-1.9086	0.5534	0.1968	-1.9086	-1.7118	0.3777	0.2894
0.5698	0.4282	800	0.5834	-4.3085	-4.8785	0.6899	0.5700	-4.8785	-4.3085	0.4424	0.3680
0.5645	0.6422	1200	0.5407	-5.3672	-6.1336	0.7196	0.7664	-6.1336	-5.3672	0.5839	0.4867
0.4723	0.8563	1600	0.5308	-6.0239	-6.7829	0.7188	0.7590	-6.7829	-6.0239	0.4449	0.3580
0.5671	1.0704	2000	0.5245	-6.1299	-6.9744	0.7270	0.8445	-6.9744	-6.1299	0.5458	0.4536
0.5184	1.2845	2400	0.5194	-6.2767	-7.2502	0.7300	0.9736	-7.2502	-6.2767	0.5423	0.4595
0.4823	1.4986	2800	0.5166	-6.4303	-7.3916	0.7285	0.9613	-7.3916	-6.4303	0.4681	0.4003
0.5627	1.7127	3200	0.5134	-6.6572	-7.6688	0.7352	1.0116	-7.6688	-6.6572	0.5174	0.4489
0.5355	1.9267	3600	0.5093	-6.3599	-7.3630	0.7352	1.0031	-7.3630	-6.3599	0.4672	0.4010
0.3968	2.1408	4000	0.5234	-7.4930	-8.6276	0.7248	1.1346	-8.6276	-7.4930	0.5678	0.5128
0.4135	2.3549	4400	0.5203	-7.4952	-8.6565	0.7240	1.1613	-8.6565	-7.4952	0.4661	0.4203
0.4277	2.5690	4800	0.5189	-7.3524	-8.5007	0.7270	1.1483	-8.5007	-7.3524	0.5701	0.5143
0.3999	2.7831	5200	0.5187	-7.4281	-8.5789	0.7292	1.1507	-8.5789	-7.4281	0.5522	0.4986
0.3855	2.9972	5600	0.5195	-7.4572	-8.6117	0.7285	1.1545	-8.6117	-7.4572	0.5435	0.4914

Framework versions

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