qwen2.5-0.5b-expo-DPO-noES4-1

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: 2.0126
• Logps: -79.3398
• Logits: -0.6948
• Objective: 1.9384
• Dpo Loss: 1.9384
• Regularize: 1.9384
• Ranking Simple: 0.5430
• Wo Beta: 6.8782

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
0.9504	0.1417	50	0.9836	-92.0961	-1.4075	0.9764	0.9764	0.9764	0.5259	7.7454
1.2453	0.2834	100	1.2755	-89.9832	-1.3843	1.2122	1.2122	1.2122	0.5336	7.3870
1.4306	0.4251	150	1.7989	-76.2439	-1.2411	1.7433	1.7433	1.7433	0.5357	7.2217
1.3076	0.5668	200	1.9202	-74.0915	-1.1859	1.8276	1.8276	1.8276	0.5388	7.0633
1.3919	0.7085	250	2.0724	-78.0833	-1.1719	2.0197	2.0197	2.0197	0.5331	7.1561
1.1295	0.8503	300	2.0572	-82.8187	-0.9338	1.9944	1.9944	1.9944	0.5404	6.9967
0.9469	0.9920	350	2.1175	-81.5331	-0.9147	1.9927	1.9927	1.9927	0.5383	6.8160
0.6125	1.1337	400	2.2348	-81.9750	-0.7963	2.1620	2.1620	2.1620	0.5342	6.9279
0.5932	1.2754	450	2.0919	-81.2039	-0.8203	2.0152	2.0152	2.0152	0.5378	6.8450
0.6283	1.4171	500	2.1655	-82.7384	-0.6531	2.0804	2.0804	2.0804	0.5393	6.7975
0.5148	1.5588	550	2.0636	-83.5332	-0.7097	1.9877	1.9877	1.9877	0.5352	6.7990
0.3647	1.7005	600	2.0351	-80.1348	-0.6773	1.9649	1.9649	1.9649	0.5373	6.8458
0.5325	1.8422	650	2.0596	-80.1766	-0.6632	1.9878	1.9878	1.9878	0.5419	6.8518
0.3442	1.9839	700	2.0875	-80.5005	-0.6144	1.9977	1.9977	1.9977	0.5435	6.7884
0.119	2.1256	750	2.0463	-80.5714	-0.6875	1.9523	1.9523	1.9523	0.5445	6.8488
0.1373	2.2674	800	2.0092	-80.3873	-0.6842	1.9315	1.9315	1.9315	0.5461	6.8647
0.131	2.4091	850	2.0060	-80.0760	-0.6729	1.9353	1.9353	1.9353	0.5450	6.8928
0.1325	2.5508	900	2.0058	-79.8102	-0.6821	1.9324	1.9324	1.9324	0.5450	6.8844
0.138	2.6925	950	2.0122	-79.3969	-0.6917	1.9363	1.9363	1.9363	0.5435	6.8712
0.1283	2.8342	1000	2.0122	-79.3139	-0.6959	1.9376	1.9376	1.9376	0.5430	6.8769
0.0938	2.9759	1050	2.0126	-79.3398	-0.6948	1.9384	1.9384	1.9384	0.5430	6.8782

Framework versions

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