IE_L3_1000steps_1e5rate_05beta_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: -1.9138
• Rewards/rejected: -16.8689
• Rewards/accuracies: 0.7400
• Rewards/margins: 14.9551
• Logps/rejected: -109.3650
• Logps/chosen: -86.6253
• Logits/rejected: -0.7926
• Logits/chosen: -0.7113

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-05
• 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
0.1906	0.4	50	0.1802	-1.6520	-15.8076	0.7400	14.1556	-107.2424	-86.1018	-0.7917	-0.7117
0.1386	0.8	100	0.1802	-1.8267	-16.5557	0.7400	14.7290	-108.7386	-86.4511	-0.7906	-0.7103
0.1386	1.2	150	0.1802	-1.8547	-16.5627	0.7400	14.7080	-108.7527	-86.5072	-0.7921	-0.7119
0.1733	1.6	200	0.1802	-1.8689	-16.5821	0.7400	14.7132	-108.7914	-86.5355	-0.7914	-0.7112
0.2253	2.0	250	0.1802	-1.8605	-16.6156	0.7400	14.7552	-108.8585	-86.5187	-0.7914	-0.7110
0.1386	2.4	300	0.1802	-1.8594	-16.6192	0.7400	14.7598	-108.8657	-86.5166	-0.7911	-0.7110
0.1213	2.8	350	0.1802	-1.8731	-16.6287	0.7400	14.7556	-108.8846	-86.5440	-0.7901	-0.7097
0.1906	3.2	400	0.1802	-1.8656	-16.7018	0.7400	14.8363	-109.0309	-86.5289	-0.7915	-0.7108
0.1906	3.6	450	0.1802	-1.8643	-16.6935	0.7400	14.8292	-109.0142	-86.5264	-0.7910	-0.7101
0.2079	4.0	500	0.1802	-1.8487	-16.6943	0.7400	14.8456	-109.0159	-86.4952	-0.7915	-0.7105
0.156	4.4	550	0.1802	-1.8609	-16.7207	0.7400	14.8598	-109.0686	-86.5195	-0.7923	-0.7110
0.1213	4.8	600	0.1802	-1.8764	-16.7597	0.7400	14.8833	-109.1467	-86.5507	-0.7921	-0.7111
0.1906	5.2	650	0.1802	-1.8747	-16.8014	0.7400	14.9267	-109.2300	-86.5471	-0.7919	-0.7103
0.2426	5.6	700	0.1802	-1.8684	-16.7797	0.7400	14.9113	-109.1867	-86.5346	-0.7925	-0.7117
0.2599	6.0	750	0.1802	-1.8981	-16.8462	0.7400	14.9481	-109.3197	-86.5939	-0.7929	-0.7119
0.1213	6.4	800	0.1802	-1.8918	-16.8690	0.7400	14.9772	-109.3652	-86.5813	-0.7929	-0.7119
0.2426	6.8	850	0.1802	-1.8689	-16.8074	0.7400	14.9386	-109.2421	-86.5355	-0.7932	-0.7122
0.1733	7.2	900	0.1802	-1.8717	-16.8482	0.7400	14.9765	-109.3236	-86.5412	-0.7924	-0.7110
0.1386	7.6	950	0.1802	-1.9143	-16.8686	0.7400	14.9543	-109.3644	-86.6264	-0.7926	-0.7113
0.156	8.0	1000	0.1802	-1.9138	-16.8689	0.7400	14.9551	-109.3650	-86.6253	-0.7926	-0.7113

Framework versions

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