This model is a fine-tuned version of tsavage68/Na_M2_1000steps_1e7_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.0000
Rewards/chosen: 2.3496
Rewards/rejected: -12.1508
Rewards/accuracies: 1.0
Rewards/margins: 14.5003
Logps/rejected: -201.4312
Logps/chosen: -24.6368
Logits/rejected: -2.3632
Logits/chosen: -2.3967

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: 200

Training results

Epoch	Step	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.2667	50	2.1916	-10.0354	1.0	12.2270	-180.2773	-26.2167	-2.3982	-2.4261
0.5333	100	2.3125	-11.6667	1.0	13.9792	-196.5901	-25.0075	-2.3692	-2.4015
0.8	150	2.3477	-12.1123	1.0	14.4600	-201.0466	-24.6557	-2.3646	-2.3980
1.0667	200	2.3496	-12.1508	1.0	14.5003	-201.4312	-24.6368	-2.3632	-2.3967

Framework versions

Transformers 4.44.2
Pytorch 2.4.0+cu121
Datasets 2.21.0
Tokenizers 0.19.1

tsavage68/Na_M2_200steps_1e6rate_01beta_cSFTDPO

Na_M2_200steps_1e7rate_01beta_cSFTDPO

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
Na_M2_200steps_1e6rate_01beta_cSFTDPO