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: 1.0416
Sft Loss: 1.4031
Rewards/chosen: -1.3990
Rewards/rejected: -1.8440
Rewards/accuracies: 0.6157
Rewards/margins: 0.4450
Logps/rejected: -1.8440
Logps/chosen: -1.3990
Logits/rejected: 0.2187
Logits/chosen: 0.1269

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
1.09	0.2141	400	1.1010	1.3681	-1.3477	-1.4855	0.5586	0.1378	-1.4855	-1.3477	0.3207	0.2350
1.0764	0.4282	800	1.0739	1.3759	-1.3603	-1.5873	0.5823	0.2270	-1.5873	-1.3603	0.3806	0.2884
1.077	0.6422	1200	1.0591	1.3822	-1.3685	-1.6704	0.5935	0.3019	-1.6704	-1.3685	0.3589	0.2649
1.0489	0.8563	1600	1.0555	1.3767	-1.3518	-1.6477	0.5905	0.2959	-1.6477	-1.3518	0.4297	0.3293
1.1366	1.0704	2000	1.0496	1.3798	-1.3555	-1.7040	0.5987	0.3484	-1.7040	-1.3555	0.3416	0.2453
1.0133	1.2845	2400	1.0461	1.3864	-1.3639	-1.7321	0.6053	0.3682	-1.7321	-1.3639	0.3701	0.2708
1.1144	1.4986	2800	1.0443	1.3887	-1.3652	-1.7447	0.6105	0.3794	-1.7447	-1.3652	0.2150	0.1278
1.0196	1.7127	3200	1.0449	1.3841	-1.3615	-1.7338	0.6142	0.3723	-1.7338	-1.3615	0.1872	0.1007
1.0023	1.9267	3600	1.0405	1.3927	-1.3767	-1.7830	0.6120	0.4063	-1.7830	-1.3767	0.2211	0.1322
0.9654	2.1408	4000	1.0418	1.3967	-1.3910	-1.8183	0.6180	0.4273	-1.8183	-1.3910	0.2405	0.1482
0.9676	2.3549	4400	1.0418	1.4054	-1.4061	-1.8540	0.6231	0.4479	-1.8540	-1.4061	0.2064	0.1158
0.9789	2.5690	4800	1.0420	1.4009	-1.3974	-1.8380	0.6142	0.4406	-1.8380	-1.3974	0.1887	0.0996
1.0003	2.7831	5200	1.0413	1.4027	-1.3986	-1.8438	0.6187	0.4452	-1.8438	-1.3986	0.2046	0.1137
0.9909	2.9972	5600	1.0416	1.4031	-1.3990	-1.8440	0.6157	0.4450	-1.8440	-1.3990	0.2187	0.1269

Framework versions

Transformers 4.44.2
Pytorch 2.2.2+cu121
Datasets 2.18.0
Tokenizers 0.19.1

RichardErkhov/yakazimir_-_qwen_cpo_entropy_0_3-awq

Original model description:

qwen_cpo_entropy_0_3

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

RichardErkhov/
yakazimir_-_qwen_cpo_entropy_0_3-awq