IE_M2_1000steps_1e6rate_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: -0.0096
• Rewards/rejected: -8.8855
• Rewards/accuracies: 0.4600
• Rewards/margins: 8.8759
• Logps/rejected: -129.8764
• Logps/chosen: -42.3012
• Logits/rejected: -2.8667
• Logits/chosen: -2.7910

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
0.4507	0.4	50	0.3744	-0.2335	-4.5985	0.4600	4.3650	-87.0066	-44.5404	-2.8777	-2.8157
0.3812	0.8	100	0.3743	-0.5108	-6.6921	0.4600	6.1813	-107.9430	-47.3138	-2.8657	-2.7955
0.3119	1.2	150	0.3743	-0.1626	-7.3145	0.4600	7.1519	-114.1667	-43.8313	-2.8595	-2.7859
0.3639	1.6	200	0.3743	-0.0733	-7.7721	0.4600	7.6988	-118.7424	-42.9385	-2.8656	-2.7905
0.4332	2.0	250	0.3743	-0.0463	-8.0479	0.4600	8.0016	-121.5008	-42.6684	-2.8656	-2.7903
0.3986	2.4	300	0.3743	-0.0312	-8.2241	0.4600	8.1929	-123.2630	-42.5179	-2.8658	-2.7905
0.3986	2.8	350	0.3743	-0.0173	-8.3343	0.4600	8.3171	-124.3653	-42.3781	-2.8664	-2.7908
0.4505	3.2	400	0.3743	-0.0158	-8.5177	0.4600	8.5019	-126.1987	-42.3632	-2.8666	-2.7910
0.4505	3.6	450	0.3743	-0.0135	-8.5518	0.4600	8.5383	-126.5393	-42.3402	-2.8666	-2.7910
0.4332	4.0	500	0.3743	-0.0117	-8.6642	0.4600	8.6525	-127.6642	-42.3228	-2.8665	-2.7909
0.3292	4.4	550	0.3743	-0.0128	-8.6957	0.4600	8.6829	-127.9786	-42.3337	-2.8666	-2.7910
0.3639	4.8	600	0.3743	-0.0122	-8.7991	0.4600	8.7869	-129.0126	-42.3276	-2.8671	-2.7915
0.4505	5.2	650	0.3743	-0.0110	-8.8312	0.4600	8.8202	-129.3338	-42.3151	-2.8667	-2.7910
0.4505	5.6	700	0.3743	-0.0140	-8.8523	0.4600	8.8383	-129.5449	-42.3457	-2.8668	-2.7911
0.3639	6.0	750	0.3743	-0.0142	-8.8760	0.4600	8.8618	-129.7817	-42.3476	-2.8666	-2.7909
0.2426	6.4	800	0.3743	-0.0114	-8.8848	0.4600	8.8734	-129.8699	-42.3197	-2.8667	-2.7910
0.5025	6.8	850	0.3743	-0.0110	-8.8824	0.4600	8.8714	-129.8454	-42.3153	-2.8666	-2.7910
0.3119	7.2	900	0.3743	-0.0122	-8.8932	0.4600	8.8810	-129.9536	-42.3276	-2.8668	-2.7911
0.3466	7.6	950	0.3743	-0.0106	-8.8884	0.4600	8.8778	-129.9054	-42.3112	-2.8667	-2.7910
0.3812	8.0	1000	0.3743	-0.0096	-8.8855	0.4600	8.8759	-129.8764	-42.3012	-2.8667	-2.7910

Framework versions

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