qwen2.5-0.5b-expo-L2EXPO-25-1

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft3-25-1 on the hZzy/train_pairwise_all_new2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2196
• Objective: 0.2162
• Reward Accuracy: 0.6247
• Logp Accuracy: 0.5190
• Log Diff Policy: 0.5202
• Chosen Logps: -91.5077
• Rejected Logps: -92.0279
• Chosen Rewards: 0.0552
• Rejected Rewards: 0.0180
• Logits: -1.3057

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-07
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 12
• total_train_batch_size: 288
• total_eval_batch_size: 24
• 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	Objective	Reward Accuracy	Logp Accuracy	Log Diff Policy	Chosen Logps	Rejected Logps	Chosen Rewards	Rejected Rewards	Logits
0.2221	0.1577	50	0.2308	0.2270	0.5649	0.5145	0.2791	-92.8045	-93.0836	-0.0745	-0.0875	-1.3046
0.215	0.3154	100	0.2272	0.2238	0.6023	0.5157	0.3836	-93.1475	-93.5311	-0.1088	-0.1323	-1.3103
0.2132	0.4731	150	0.2243	0.2213	0.6012	0.5168	0.4200	-91.9704	-92.3904	0.0089	-0.0182	-1.3101
0.1848	0.6307	200	0.2223	0.2194	0.6096	0.5145	0.4527	-92.7829	-93.2356	-0.0723	-0.1027	-1.3491
0.1831	0.7884	250	0.2239	0.2207	0.6107	0.5190	0.4721	-91.7648	-92.2369	0.0295	-0.0028	-1.2826
0.1723	0.9461	300	0.2213	0.2186	0.6225	0.5173	0.4962	-91.0092	-91.5054	0.1050	0.0703	-1.3274
0.1411	1.1038	350	0.2244	0.2209	0.6119	0.5185	0.5052	-92.0883	-92.5935	-0.0029	-0.0385	-1.3190
0.1412	1.2615	400	0.2208	0.2173	0.6130	0.5179	0.5072	-90.6094	-91.1167	0.1450	0.1092	-1.3359
0.1309	1.4192	450	0.2207	0.2173	0.6281	0.5201	0.5157	-91.4469	-91.9627	0.0613	0.0246	-1.3163
0.1336	1.5769	500	0.2212	0.2179	0.6219	0.5190	0.5056	-91.5764	-92.0820	0.0483	0.0126	-1.3171
0.1203	1.7346	550	0.2207	0.2176	0.6320	0.5196	0.5051	-91.1888	-91.6939	0.0871	0.0515	-1.3031
0.1193	1.8922	600	0.2207	0.2173	0.6275	0.5173	0.5090	-91.7077	-92.2167	0.0352	-0.0008	-1.2903
0.0974	2.0499	650	0.2198	0.2156	0.6320	0.5179	0.5221	-91.5659	-92.0880	0.0494	0.0120	-1.3050
0.0946	2.2076	700	0.2201	0.2163	0.6342	0.5196	0.5182	-91.6268	-92.1451	0.0433	0.0063	-1.2980
0.0959	2.3653	750	0.2195	0.2163	0.6370	0.5179	0.5202	-91.4802	-92.0004	0.0580	0.0208	-1.3090
0.0932	2.5230	800	0.2197	0.2165	0.6236	0.5196	0.5172	-91.5272	-92.0445	0.0532	0.0164	-1.3048
0.0904	2.6807	850	0.2196	0.2163	0.6270	0.5190	0.5193	-91.5233	-92.0425	0.0536	0.0166	-1.3060
0.0904	2.8384	900	0.2195	0.2162	0.6258	0.5190	0.5202	-91.5277	-92.0479	0.0532	0.0161	-1.3057
0.0963	2.9961	950	0.2196	0.2162	0.6247	0.5190	0.5202	-91.5078	-92.0279	0.0552	0.0180	-1.3057

Framework versions

• Transformers 4.42.0
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.19.1