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

Loss: 0.0001
Logps: -91.2444
Logits: -1.4897
Objective: 0.0001
Dpo Loss: 0.6779
Regularize: 0.4017
Ranking Simple: 0.5316
Ranking Idealized: 0.6025
Ranking Idealized Expo: 0.5233
Wo Beta: 16.5542

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: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 3
gradient_accumulation_steps: 12
total_train_batch_size: 144
total_eval_batch_size: 12
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

Training results

Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo	Wo Beta
0.1417	50	0.0001	-90.6039	-1.4633	0.0001	0.6886	0.4141	0.5243	0.6025	0.5233	16.5848
0.2834	100	0.0001	-91.6434	-1.5468	0.0001	0.6789	0.4080	0.5285	0.6025	0.5233	16.2910
0.4251	150	0.0001	-92.9071	-1.4692	0.0001	0.6790	0.4111	0.5331	0.6025	0.5233	16.8081
0.5668	200	0.0001	-94.1014	-1.5088	0.0001	0.6764	0.4073	0.5280	0.6025	0.5233	16.5494
0.7085	250	0.0001	-92.4359	-1.5576	0.0001	0.6787	0.4095	0.5326	0.6025	0.5233	16.7036
0.8503	300	0.0001	-91.3976	-1.4988	0.0001	0.6777	0.4058	0.5305	0.6025	0.5233	16.5040
0.9920	350	0.0001	-92.9950	-1.4920	0.0001	0.6792	0.4098	0.5305	0.6025	0.5233	16.7141
1.1337	400	0.0001	-90.4269	-1.4778	0.0001	0.6770	0.4009	0.5347	0.6025	0.5233	16.6248
1.2754	450	0.0001	-92.0877	-1.4925	0.0001	0.6797	0.4066	0.5378	0.6025	0.5233	16.5429
1.4171	500	0.0001	-89.2338	-1.4505	0.0001	0.6779	0.4022	0.5357	0.6025	0.5233	16.5511
1.5588	550	0.0001	-90.7047	-1.4772	0.0001	0.6778	0.4019	0.5342	0.6025	0.5233	16.4827
1.7005	600	0.0001	-90.5059	-1.4760	0.0001	0.6775	0.4020	0.5352	0.6025	0.5233	16.4456
1.8422	650	0.0001	-90.5418	-1.4723	0.0001	0.6776	0.4024	0.5321	0.6025	0.5233	16.5562
1.9839	700	0.0001	-90.7432	-1.4788	0.0001	0.6785	0.4029	0.5331	0.6025	0.5233	16.5133
2.1256	750	0.0001	-91.2051	-1.4918	0.0001	0.6773	0.4014	0.5336	0.6025	0.5233	16.5438
2.2674	800	0.0001	-91.2034	-1.4901	0.0001	0.6774	0.4009	0.5326	0.6025	0.5233	16.5337
2.4091	850	0.0001	-91.0159	-1.4877	0.0001	0.6778	0.4018	0.5331	0.6025	0.5233	16.5458
2.5508	900	0.0001	-91.1343	-1.4912	0.0001	0.6779	0.4018	0.5321	0.6025	0.5233	16.5533
2.6925	950	0.0001	-91.2303	-1.4921	0.0001	0.6779	0.4018	0.5316	0.6025	0.5233	16.5485
2.8342	1000	0.0001	-91.2246	-1.4900	0.0001	0.6779	0.4017	0.5316	0.6025	0.5233	16.5531
2.9759	1050	0.0001	-91.2444	-1.4897	0.0001	0.6779	0.4017	0.5316	0.6025	0.5233	16.5542

Framework versions

Transformers 4.42.0
Pytorch 2.3.0+cu121
Datasets 3.2.0
Tokenizers 0.19.1

hZzy/qwen2.5-0.5b-expo-L2EXPO-W1-noES-0.1

qwen2.5-0.5b-expo-L2EXPO-W1-noES-0.1

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

hZzy/
qwen2.5-0.5b-expo-L2EXPO-W1-noES-0.1