IE_M2_1000steps_1e7rate_01beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3743
• Rewards/chosen: -0.3291
• Rewards/rejected: -6.1017
• Rewards/accuracies: 0.4600
• Rewards/margins: 5.7727
• Logps/rejected: -102.0393
• Logps/chosen: -45.4965
• Logits/rejected: -2.8684
• Logits/chosen: -2.8050

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-07
• 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
0.558	0.4	50	0.4553	-0.0349	-0.8002	0.4600	0.7653	-49.0237	-42.5545	-2.9038	-2.8422
0.3818	0.8	100	0.3747	-0.1730	-3.5887	0.4600	3.4157	-76.9091	-43.9359	-2.8759	-2.8145
0.3123	1.2	150	0.3744	-0.2403	-4.3676	0.4600	4.1273	-84.6980	-44.6088	-2.8742	-2.8132
0.364	1.6	200	0.3744	-0.2016	-4.5800	0.4600	4.3784	-86.8216	-44.2215	-2.8745	-2.8130
0.4332	2.0	250	0.3743	-0.2684	-4.8731	0.4600	4.6046	-89.7525	-44.8898	-2.8737	-2.8118
0.3986	2.4	300	0.3743	-0.1931	-5.0362	0.4600	4.8430	-91.3835	-44.1367	-2.8747	-2.8125
0.3986	2.8	350	0.3743	-0.1846	-5.1505	0.4600	4.9659	-92.5268	-44.0517	-2.8745	-2.8120
0.4506	3.2	400	0.3743	-0.1881	-5.2928	0.4600	5.1047	-93.9497	-44.0868	-2.8736	-2.8107
0.4505	3.6	450	0.3743	-0.2250	-5.5587	0.4600	5.3337	-96.6092	-44.4557	-2.8724	-2.8094
0.4332	4.0	500	0.3743	-0.4284	-5.9879	0.4600	5.5595	-100.9007	-46.4892	-2.8698	-2.8066
0.3292	4.4	550	0.3743	-0.3669	-5.9892	0.4600	5.6223	-100.9135	-45.8741	-2.8695	-2.8063
0.3639	4.8	600	0.3743	-0.2855	-5.9594	0.4600	5.6739	-100.6163	-45.0607	-2.8699	-2.8066
0.4505	5.2	650	0.3743	-0.3591	-6.0896	0.4600	5.7305	-101.9183	-45.7970	-2.8685	-2.8052
0.4505	5.6	700	0.3743	-0.3292	-6.0868	0.4600	5.7576	-101.8900	-45.4977	-2.8687	-2.8054
0.3639	6.0	750	0.3743	-0.3284	-6.1008	0.4600	5.7724	-102.0299	-45.4898	-2.8683	-2.8049
0.2426	6.4	800	0.3743	-0.3283	-6.0983	0.4600	5.7700	-102.0044	-45.4881	-2.8684	-2.8051
0.5025	6.8	850	0.3743	-0.3251	-6.0987	0.4600	5.7737	-102.0092	-45.4562	-2.8685	-2.8051
0.3119	7.2	900	0.3743	-0.3297	-6.1009	0.4600	5.7712	-102.0308	-45.5028	-2.8684	-2.8050
0.3466	7.6	950	0.3743	-0.3291	-6.1017	0.4600	5.7727	-102.0393	-45.4965	-2.8684	-2.8050
0.3812	8.0	1000	0.3743	-0.3291	-6.1017	0.4600	5.7727	-102.0393	-45.4965	-2.8684	-2.8050

Framework versions

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