IE_L3_1000steps_1e6rate_03beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1802
• Rewards/chosen: -1.3199
• Rewards/rejected: -13.3530
• Rewards/accuracies: 0.7400
• Rewards/margins: 12.0331
• Logps/rejected: -120.1372
• Logps/chosen: -87.1973
• Logits/rejected: -0.8052
• Logits/chosen: -0.7124

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: 1e-06
• train_batch_size: 2
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.1907	0.4	50	0.1802	-1.0923	-10.4680	0.7400	9.3757	-110.5205	-86.4386	-0.7963	-0.7114
0.1386	0.8	100	0.1802	-1.2190	-11.5716	0.7400	10.3526	-114.1993	-86.8611	-0.7960	-0.7088
0.1386	1.2	150	0.1802	-1.2269	-11.8797	0.7400	10.6528	-115.2263	-86.8875	-0.7973	-0.7092
0.1733	1.6	200	0.1802	-1.2628	-12.4562	0.7400	11.1934	-117.1479	-87.0072	-0.7983	-0.7088
0.2253	2.0	250	0.1802	-1.2811	-12.6109	0.7400	11.3298	-117.6637	-87.0682	-0.8005	-0.7100
0.1386	2.4	300	0.1802	-1.2819	-12.6821	0.7400	11.4002	-117.9011	-87.0709	-0.8009	-0.7104
0.1213	2.8	350	0.1802	-1.2857	-12.9252	0.7400	11.6395	-118.7114	-87.0834	-0.8024	-0.7110
0.1906	3.2	400	0.1802	-1.2904	-12.9929	0.7400	11.7024	-118.9368	-87.0992	-0.8026	-0.7109
0.1906	3.6	450	0.1802	-1.2935	-13.0320	0.7400	11.7385	-119.0673	-87.1095	-0.8030	-0.7112
0.2079	4.0	500	0.1802	-1.3034	-13.1728	0.7400	11.8694	-119.5364	-87.1423	-0.8047	-0.7126
0.156	4.4	550	0.1802	-1.3085	-13.2242	0.7400	11.9157	-119.7078	-87.1593	-0.8035	-0.7118
0.1213	4.8	600	0.1802	-1.2992	-13.2411	0.7400	11.9418	-119.7642	-87.1285	-0.8054	-0.7131
0.1906	5.2	650	0.1802	-1.3144	-13.3156	0.7400	12.0011	-120.0125	-87.1792	-0.8048	-0.7117
0.2426	5.6	700	0.1802	-1.2925	-13.3031	0.7400	12.0106	-119.9710	-87.1061	-0.8043	-0.7117
0.2599	6.0	750	0.1802	-1.3084	-13.3298	0.7400	12.0213	-120.0597	-87.1592	-0.8052	-0.7126
0.1213	6.4	800	0.1802	-1.3118	-13.3477	0.7400	12.0359	-120.1197	-87.1704	-0.8039	-0.7116
0.2426	6.8	850	0.1802	-1.3228	-13.3620	0.7400	12.0392	-120.1673	-87.2071	-0.8052	-0.7125
0.1733	7.2	900	0.1802	-1.3137	-13.3379	0.7400	12.0242	-120.0870	-87.1768	-0.8052	-0.7125
0.1386	7.6	950	0.1802	-1.3070	-13.3530	0.7400	12.0460	-120.1374	-87.1545	-0.8053	-0.7127
0.156	8.0	1000	0.1802	-1.3199	-13.3530	0.7400	12.0331	-120.1372	-87.1973	-0.8052	-0.7124

Framework versions

• Transformers 4.44.2
• Pytorch 2.0.0+cu117
• Datasets 3.0.0
• Tokenizers 0.19.1