qwen2.5-0.5b-expo-L2EXPO-ES-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: 0.6458
• Logps: -79.3784
• Logits: -0.5665
• Objective: 0.6297
• Dpo Loss: 0.7206
• Regularize: 0.6297
• Ranking Simple: 0.5347
• Ranking Idealized: 0.6030
• Ranking Idealized Expo: 0.5223
• Wo Beta: 14.3163

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo	Wo Beta
0.4017	0.1417	50	0.4165	-93.1726	-1.5024	0.4149	0.6868	0.4149	0.5259	0.6030	0.5223	16.4267
0.3777	0.2834	100	0.4360	-92.8653	-1.4775	0.4269	0.6818	0.4269	0.5316	0.6030	0.5223	16.2439
0.4057	0.4251	150	0.4911	-84.1774	-1.2946	0.4805	0.6897	0.4805	0.5383	0.6030	0.5223	15.6306
0.4475	0.5668	200	0.5660	-89.7342	-0.9897	0.5515	0.7103	0.5515	0.5316	0.6030	0.5223	15.1280
0.455	0.7085	250	0.5978	-78.1917	-1.0033	0.5822	0.7171	0.5822	0.5311	0.6030	0.5223	14.6763
0.4337	0.8503	300	0.5993	-78.8918	-0.6761	0.5779	0.7105	0.5779	0.5300	0.6030	0.5223	14.9196
0.4039	0.9920	350	0.5978	-75.1520	-0.7968	0.5765	0.7078	0.5765	0.5290	0.6030	0.5223	14.6531
0.3729	1.1337	400	0.6180	-75.1433	-0.5569	0.6000	0.7153	0.6000	0.5228	0.6030	0.5223	14.6471
0.3454	1.2754	450	0.6316	-76.2289	-0.6214	0.6131	0.7165	0.6131	0.5336	0.6030	0.5223	14.5034
0.3226	1.4171	500	0.6255	-77.6040	-0.5608	0.6084	0.7204	0.6084	0.5285	0.6030	0.5223	14.4998
0.3133	1.5588	550	0.6282	-78.6291	-0.6736	0.6138	0.7139	0.6138	0.5336	0.6030	0.5223	14.4069
0.2944	1.7005	600	0.6321	-78.9179	-0.5620	0.6139	0.7175	0.6139	0.5357	0.6030	0.5223	14.6142
0.2915	1.8422	650	0.6321	-77.4437	-0.7021	0.6157	0.7138	0.6157	0.5367	0.6030	0.5223	14.3858
0.2675	1.9839	700	0.6386	-79.3600	-0.5612	0.6233	0.7185	0.6233	0.5290	0.6030	0.5223	14.3171
0.2415	2.1256	750	0.6405	-80.0990	-0.6174	0.6263	0.7177	0.6263	0.5347	0.6030	0.5223	14.4302
0.2263	2.2674	800	0.6458	-79.3784	-0.5665	0.6297	0.7206	0.6297	0.5347	0.6030	0.5223	14.3163
0.2148	2.4091	850	0.6436	-79.0806	-0.5793	0.6276	0.7192	0.6276	0.5362	0.6030	0.5223	14.4263
0.1993	2.5508	900	0.6454	-80.3815	-0.5621	0.6302	0.7217	0.6302	0.5342	0.6030	0.5223	14.4491
0.1887	2.6925	950	0.6443	-79.1446	-0.6216	0.6274	0.7204	0.6274	0.5336	0.6030	0.5223	14.3186
0.1764	2.8342	1000	0.6399	-79.7721	-0.6087	0.6246	0.7200	0.6246	0.5336	0.6030	0.5223	14.4502
0.163	2.9759	1050	0.6428	-79.5818	-0.6068	0.6266	0.7211	0.6266	0.5316	0.6030	0.5223	14.3406

Framework versions

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