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

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: 1.3212
• Logps: -78.9331
• Logits: -0.6046
• Objective: 1.2922
• Dpo Loss: 0.7096
• Regularize: 0.5826
• Ranking Simple: 0.5357

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
1.0668	0.1417	50	1.1061	-90.1942	-1.4829	1.1093	0.6857	0.4236	0.5300
1.0269	0.2834	100	1.1677	-92.1258	-1.4409	1.1407	0.6807	0.4600	0.5326
1.0823	0.4251	150	1.2316	-81.0954	-1.3424	1.2139	0.6936	0.5204	0.5285
1.0577	0.5668	200	1.2470	-82.5828	-1.0627	1.2263	0.6939	0.5324	0.5331
0.9955	0.7085	250	1.2858	-81.0205	-1.0947	1.2580	0.7011	0.5569	0.5347
0.954	0.8503	300	1.2727	-82.4789	-0.9186	1.2483	0.6948	0.5535	0.5409
0.9014	0.9920	350	1.2973	-80.3169	-0.8042	1.2672	0.7021	0.5651	0.5362
0.8458	1.1337	400	1.3161	-78.6994	-0.5803	1.2922	0.7089	0.5833	0.5383
0.8325	1.2754	450	1.3054	-79.2087	-0.6878	1.2837	0.7065	0.5772	0.5378
0.796	1.4171	500	1.3290	-79.5455	-0.6465	1.3067	0.7132	0.5934	0.5383
0.7784	1.5588	550	1.3215	-78.1244	-0.6049	1.2954	0.7083	0.5871	0.5414
0.753	1.7005	600	1.3166	-78.2126	-0.5817	1.2870	0.7062	0.5808	0.5373
0.738	1.8422	650	1.3141	-78.5070	-0.6067	1.2850	0.7055	0.5794	0.5378
0.7128	1.9839	700	1.3177	-78.7581	-0.6380	1.2901	0.7085	0.5816	0.5404
0.6518	2.1256	750	1.3227	-79.7230	-0.6694	1.2915	0.7083	0.5832	0.5393
0.6576	2.2674	800	1.3182	-79.6166	-0.6259	1.2881	0.7079	0.5801	0.5367
0.6373	2.4091	850	1.3180	-79.0937	-0.5955	1.2881	0.7083	0.5798	0.5367
0.6338	2.5508	900	1.3195	-78.9960	-0.6047	1.2905	0.7091	0.5814	0.5347
0.631	2.6925	950	1.3215	-78.8580	-0.6057	1.2924	0.7096	0.5828	0.5362
0.6254	2.8342	1000	1.3215	-78.9196	-0.6057	1.2924	0.7096	0.5828	0.5373
0.6187	2.9759	1050	1.3212	-78.9331	-0.6046	1.2922	0.7096	0.5826	0.5357

Framework versions

• Transformers 4.42.0
• Pytorch 2.3.0+cu121
• Datasets 3.2.0
• Tokenizers 0.19.1