This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-news-IFT on the hZzy/train_pairwise dataset. It achieves the following results on the evaluation set:

Loss: 1.0827
Logps: -99.3608
Logits: -1.3686
Objective: 1.0902
Dpo Loss: 0.6830
Regularize: 0.4072
Ranking Simple: 0.5221
Ranking Idealized: 0.5925
Ranking Idealized Expo: 0.5166

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-07
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 12
total_train_batch_size: 192
total_eval_batch_size: 16
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
1.0911	0.1889	50	1.1043	-97.7941	-1.3092	1.1042	0.6905	0.4136	0.5166	0.5925	0.5166
1.0629	0.3778	100	1.0938	-98.1420	-1.3221	1.0948	0.6871	0.4077	0.5180	0.5925	0.5166
1.0597	0.5668	150	1.0886	-97.5760	-1.3388	1.0899	0.6843	0.4057	0.5193	0.5925	0.5166
1.0201	0.7557	200	1.0878	-98.5525	-1.3519	1.0928	0.6845	0.4083	0.5193	0.5925	0.5166
1.0173	0.9446	250	1.0863	-99.3374	-1.3586	1.0919	0.6845	0.4074	0.5193	0.5925	0.5166
0.9755	1.1335	300	1.0829	-99.1144	-1.3663	1.0893	0.6826	0.4067	0.5214	0.5925	0.5166
0.9708	1.3224	350	1.0829	-99.2669	-1.3642	1.0901	0.6833	0.4068	0.5221	0.5925	0.5166
0.968	1.5113	400	1.0829	-99.2520	-1.3682	1.0901	0.6832	0.4069	0.5214	0.5925	0.5166
0.9495	1.7003	450	1.0826	-99.3216	-1.3688	1.0900	0.6830	0.4070	0.5214	0.5925	0.5166
0.9463	1.8892	500	1.0827	-99.3605	-1.3686	1.0902	0.6830	0.4072	0.5221	0.5925	0.5166

Framework versions

Transformers 4.42.0
Pytorch 2.3.0+cu121
Datasets 2.19.1
Tokenizers 0.19.1

hZzy/qwen2.5-0.5b-expo-EXDPO2

qwen2.5-0.5b-expo-EXDPO2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

hZzy/
qwen2.5-0.5b-expo-EXDPO2