UTI2_M2_50steps_1e6rate_01beta_CSFTDPO

This model is a fine-tuned version of tsavage68/UTI_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5476
• Rewards/chosen: -0.0191
• Rewards/rejected: -2.0342
• Rewards/accuracies: 0.2100
• Rewards/margins: 2.0151
• Logps/rejected: -29.7161
• Logps/chosen: -4.7332
• Logits/rejected: -2.5508
• Logits/chosen: -2.5517

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: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6238	0.3333	25	0.6989	0.0094	-1.0339	0.1900	1.0433	-19.7134	-4.4489	-2.5742	-2.5732
0.6671	0.6667	50	0.5476	-0.0191	-2.0342	0.2100	2.0151	-29.7161	-4.7332	-2.5508	-2.5517

Framework versions

• Transformers 4.41.2
• Pytorch 2.0.0+cu117
• Datasets 2.19.2
• Tokenizers 0.19.1