This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the lightblue_kto_data dataset. It achieves the following results on the evaluation set:

Loss: 0.3031
Rewards/chosen: 1.5421
Logps/chosen: -343.9051
Logits/chosen: -69679219.2
Rewards/rejected: -7.3046
Logps/rejected: -233.7684
Logits/rejected: -34451756.1379
Rewards/margins: 8.8467
Kl: 1080.3173

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: 5e-06
train_batch_size: 1
eval_batch_size: 1
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 16
total_train_batch_size: 128
total_eval_batch_size: 8
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.01
num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Logps/chosen	Logits/chosen	Rewards/rejected	Logps/rejected	Logits/rejected	Rewards/margins
0.2623	0.0997	36	0.3340	1.3847	-345.4796	-55713169.0667	-3.6384	-197.1070	-40055004.6897	5.0231	890.2159
0.3222	0.1995	72	0.3273	1.5219	-344.1068	-61469499.7333	-4.9277	-209.9999	-32503238.6207	6.4496	1189.5447
0.3798	0.2992	108	0.3185	1.5573	-343.7531	-63003302.4	-5.7081	-217.8038	-31597484.1379	7.2654	955.4995
0.3755	0.3990	144	0.3016	0.8908	-350.4181	-63924428.8	-6.8986	-229.7092	-27711788.1379	7.7895	705.8951
0.3454	0.4987	180	0.3053	1.4481	-344.8449	-67193476.2667	-6.5311	-226.0336	-37107747.3103	7.9792	836.6326
0.2633	0.5984	216	0.3085	1.5864	-343.4627	-68801646.9333	-6.4654	-225.3766	-37986458.4828	8.0517	974.3778
0.2519	0.6982	252	0.3109	1.5635	-343.6908	-69407142.4	-6.4303	-225.0262	-34758311.7241	7.9939	1106.7635
0.2959	0.7979	288	0.3033	1.6631	-342.6956	-69444923.7333	-7.0061	-230.7837	-36029797.5172	8.6691	1082.5067
0.2921	0.8977	324	0.3022	1.4322	-345.0042	-69711099.7333	-7.5841	-236.5635	-35742644.9655	9.0163	1047.6223
0.3122	0.9974	360	0.3031	1.5421	-343.9051	-69679219.2	-7.3046	-233.7684	-34451756.1379	8.8467	1080.3173

Framework versions

Transformers 4.46.1
Pytorch 2.4.0+cu121
Datasets 3.1.0
Tokenizers 0.20.3

RichardErkhov/lightblue_-_qwen2.5-7B-instruct-kto-4bits

Original model description:

kto_trained_1

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

RichardErkhov/
lightblue_-_qwen2.5-7B-instruct-kto-4bits