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: 0.8210
Logps: -136.0435
Logits: -1.9089
Objective: 0.8326
Dpo Loss: 0.8326
Regularize: 0.8326
Ranking Simple: 0.5631
Wo Beta: 10.0450

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: 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

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Wo Beta
0.644	0.1417	50	0.6810	-88.9389	-1.5667	0.6836	0.6836	0.6836	0.5316	7.8652
0.5968	0.2834	100	0.6815	-101.2053	-1.7411	0.6847	0.6847	0.6847	0.5393	7.7021
0.5144	0.4251	150	0.6850	-94.4351	-1.7346	0.6782	0.6782	0.6782	0.5549	7.3585
0.4808	0.5668	200	0.7037	-103.6052	-1.8068	0.7036	0.7036	0.7036	0.5569	7.7326
0.4863	0.7085	250	0.7026	-91.8159	-1.9767	0.6984	0.6984	0.6984	0.5476	7.8312
0.4389	0.8503	300	0.6993	-105.6110	-2.0810	0.6947	0.6947	0.6947	0.5600	7.5894
0.3851	0.9920	350	0.7227	-103.2476	-2.0184	0.7155	0.7155	0.7155	0.5492	7.9656
0.2556	1.1337	400	0.7344	-109.3563	-1.9806	0.7314	0.7314	0.7314	0.5445	8.6228
0.264	1.2754	450	0.7229	-110.4481	-1.8432	0.7204	0.7204	0.7204	0.5580	8.5473
0.2767	1.4171	500	0.7313	-111.4522	-1.9699	0.7300	0.7300	0.7300	0.5497	8.5441
0.2273	1.5588	550	0.7207	-116.6543	-1.7731	0.7313	0.7313	0.7313	0.5575	8.5606
0.2232	1.7005	600	0.7356	-115.8618	-1.7360	0.7399	0.7399	0.7399	0.5719	8.7758
0.2623	1.8422	650	0.7370	-117.7434	-2.0182	0.7381	0.7381	0.7381	0.5745	8.7274
0.2194	1.9839	700	0.7433	-121.1650	-1.9499	0.7528	0.7528	0.7528	0.5657	9.0270
0.1094	2.1256	750	0.8255	-134.3582	-1.8660	0.8363	0.8363	0.8363	0.5611	10.1139
0.1222	2.2674	800	0.8124	-133.2139	-1.9092	0.8237	0.8237	0.8237	0.5652	9.8993
0.1161	2.4091	850	0.8204	-134.1696	-1.8946	0.8314	0.8314	0.8314	0.5642	9.9670
0.1268	2.5508	900	0.8157	-135.2029	-1.8941	0.8271	0.8271	0.8271	0.5642	9.9596
0.1263	2.6925	950	0.8189	-135.8437	-1.9048	0.8305	0.8305	0.8305	0.5642	10.0013
0.1197	2.8342	1000	0.8205	-135.9884	-1.9072	0.8320	0.8320	0.8320	0.5626	10.0373
0.1192	2.9759	1050	0.8210	-136.0435	-1.9089	0.8326	0.8326	0.8326	0.5631	10.0450

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-DPO-noES2-0.1

qwen2.5-0.5b-expo-DPO-noES2-0.1

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

hZzy/
qwen2.5-0.5b-expo-DPO-noES2-0.1