qwen2.5-0.5b-expo-L2EXPO-EXPERIMENT-0.1-5e7

This model is a fine-tuned version of hZzy/qwen2.5-0.5b-sft-news-IFT on the hZzy/train_pairwise dataset. It achieves the following results on the evaluation set:

• Loss: 0.4061
• Logps: -93.6123
• Logits: -1.5421
• Objective: 0.4082
• Dpo Loss: 0.6835
• Regularize: 0.4082
• Ranking Simple: 0.5217
• Ranking Idealized: 0.5888
• Ranking Idealized Expo: 0.5103

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Logps	Logits	Objective	Dpo Loss	Regularize	Ranking Simple	Ranking Idealized	Ranking Idealized Expo
0.4009	0.2834	50	0.4075	-91.1598	-1.4550	0.4075	0.6900	0.4075	0.5145	0.5888	0.5103
0.3473	0.5668	100	0.4035	-92.0013	-1.4828	0.4074	0.6864	0.4074	0.5145	0.5888	0.5103
0.2868	0.8503	150	0.4047	-93.0470	-1.4826	0.4111	0.6868	0.4111	0.5207	0.5888	0.5103
0.2165	1.1337	200	0.4073	-93.5155	-1.5000	0.4126	0.6870	0.4126	0.5165	0.5888	0.5103
0.2005	1.4171	250	0.4063	-93.3534	-1.5170	0.4069	0.6846	0.4069	0.5196	0.5888	0.5103
0.1818	1.7005	300	0.4041	-92.4353	-1.5152	0.4060	0.6846	0.4060	0.5217	0.5888	0.5103
0.165	1.9839	350	0.4057	-93.2243	-1.5221	0.4057	0.6835	0.4057	0.5207	0.5888	0.5103
0.1389	2.2674	400	0.4073	-93.7396	-1.5308	0.4090	0.6834	0.4090	0.5176	0.5888	0.5103
0.1152	2.5508	450	0.4049	-93.8421	-1.5304	0.4060	0.6827	0.4060	0.5217	0.5888	0.5103
0.11	2.8342	500	0.4061	-93.4569	-1.5372	0.4082	0.6844	0.4082	0.5165	0.5888	0.5103
0.0961	3.1176	550	0.4054	-93.4277	-1.5393	0.4072	0.6832	0.4072	0.5186	0.5888	0.5103
0.0793	3.4010	600	0.4061	-93.4066	-1.5455	0.4071	0.6832	0.4071	0.5217	0.5888	0.5103
0.0715	3.6845	650	0.4060	-93.5656	-1.5409	0.4077	0.6834	0.4077	0.5227	0.5888	0.5103
0.0763	3.9679	700	0.4063	-93.5022	-1.5427	0.4085	0.6837	0.4085	0.5186	0.5888	0.5103
0.0623	4.2513	750	0.4061	-93.5738	-1.5419	0.4081	0.6836	0.4081	0.5227	0.5888	0.5103
0.0655	4.5347	800	0.4060	-93.5684	-1.5425	0.4082	0.6835	0.4082	0.5227	0.5888	0.5103
0.0601	4.8181	850	0.4061	-93.6119	-1.5421	0.4082	0.6835	0.4082	0.5217	0.5888	0.5103

Framework versions

• Transformers 4.42.0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1