qwen_qfUNL_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: 0.6510
• Rewards/chosen: -1.7989
• Rewards/rejected: -2.5830
• Rewards/accuracies: 0.6736
• Rewards/margins: 0.7841
• Logps/rejected: -2.5830
• Logps/chosen: -1.7989
• Logits/rejected: 0.0192
• Logits/chosen: -0.0604

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
0.6781	0.2141	400	0.6873	-1.6444	-1.8233	0.5475	0.1789	-1.8233	-1.6444	0.2857	0.1996
0.6757	0.4282	800	0.6641	-1.6348	-1.9815	0.6239	0.3467	-1.9815	-1.6348	0.3665	0.2730
0.6569	0.6422	1200	0.6602	-1.7060	-2.1644	0.6424	0.4584	-2.1644	-1.7060	0.2601	0.1749
0.6562	0.8563	1600	0.6584	-1.8368	-2.3836	0.6513	0.5468	-2.3836	-1.8368	0.1796	0.0944
0.6883	1.0704	2000	0.6545	-1.7098	-2.2986	0.6639	0.5888	-2.2986	-1.7098	0.2146	0.1248
0.6581	1.2845	2400	0.6533	-1.7444	-2.3861	0.6691	0.6417	-2.3861	-1.7444	0.1530	0.0644
0.6444	1.4986	2800	0.6537	-1.7815	-2.4833	0.6684	0.7018	-2.4833	-1.7815	0.0665	-0.0145
0.6575	1.7127	3200	0.6520	-1.7922	-2.5114	0.6654	0.7192	-2.5114	-1.7922	0.1107	0.0260
0.6481	1.9267	3600	0.6507	-1.7358	-2.4632	0.6736	0.7275	-2.4632	-1.7358	0.0939	0.0113
0.607	2.1408	4000	0.6506	-1.7686	-2.5161	0.6751	0.7475	-2.5161	-1.7686	0.0842	0.0005
0.6294	2.3549	4400	0.6514	-1.8215	-2.5986	0.6714	0.7771	-2.5986	-1.8215	0.0008	-0.0778
0.6098	2.5690	4800	0.6507	-1.7918	-2.5693	0.6766	0.7775	-2.5693	-1.7918	0.0735	-0.0103
0.6302	2.7831	5200	0.6507	-1.7943	-2.5780	0.6751	0.7837	-2.5780	-1.7943	0.0395	-0.0418
0.6181	2.9972	5600	0.6510	-1.7989	-2.5830	0.6736	0.7841	-2.5830	-1.7989	0.0192	-0.0604

Framework versions

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