Modelpublic

tsavage68/Transaminitis_L3_125steps_1e5rate_05beta_CSFTDPO

sourceHugging Facellama3updated 2y agoView on Hugging Face

0likes12downloads

Model Card

TransaminitisL3125steps1e5rate05beta_CSFTDPO

This model is a fine-tuned version of tsavage68/Transaminitis_L3_1000rate_1e7_SFT on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.7424
—Rewards/chosen: -16.3012
—Rewards/rejected: -16.2112
—Rewards/accuracies: 0.3500
—Rewards/margins: -0.0899
—Logps/rejected: -50.9772
—Logps/chosen: -51.1365
—Logits/rejected: -1.0737
—Logits/chosen: -1.0737

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
—trainbatchsize: 2
—evalbatchsize: 1
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 4
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: cosine
—lrschedulerwarmup_steps: 100
—training_steps: 125

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
1.3308	0.2	25	1.4218	-5.1457	-5.2961	0.5400	0.1503	-29.1468	-28.8257	-0.7892	-0.7880
1.1498	0.4	50	0.7304	-4.8999	-4.8425	0.4000	-0.0574	-28.2397	-28.3340	-2.1796	-2.1797
1.2832	0.6	75	0.9255	-1.6896	-4.2819	0.6300	2.5923	-27.1184	-21.9133	-1.0885	-1.0850
2.8764	0.8	100	3.8444	-19.0391	-19.6042	0.5400	0.5651	-57.7631	-56.6124	-0.1327	-0.1327
0.7767	1.0	125	0.7424	-16.3012	-16.2112	0.3500	-0.0899	-50.9772	-51.1365	-1.0737	-1.0737

Framework versions

—Transformers 4.40.2
—Pytorch 2.0.0+cu117
—Datasets 2.19.1
—Tokenizers 0.19.1