UTI_L3_150steps_1e7rate_05beta_CSFTDPO

This model is a fine-tuned version of tsavage68/UTI_L3_1000steps_1e5rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.2407
• Rewards/chosen: 0.3257
• Rewards/rejected: -1.3197
• Rewards/accuracies: 0.9900
• Rewards/margins: 1.6454
• Logps/rejected: -65.8341
• Logps/chosen: -31.8276
• Logits/rejected: -1.3243
• Logits/chosen: -1.3092

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-07
• 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: 150

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6936	0.3333	25	0.6887	-0.0007	-0.0109	0.5300	0.0102	-63.2166	-32.4804	-1.3231	-1.3079
0.6544	0.6667	50	0.6317	0.0212	-0.1104	0.8500	0.1316	-63.4155	-32.4367	-1.3229	-1.3078
0.5537	1.0	75	0.5050	0.0832	-0.3693	0.9400	0.4525	-63.9333	-32.3125	-1.3234	-1.3083
0.3404	1.3333	100	0.3397	0.1907	-0.8797	0.9800	1.0704	-64.9540	-32.0976	-1.3239	-1.3089
0.2243	1.6667	125	0.2475	0.3167	-1.2899	0.9900	1.6066	-65.7744	-31.8456	-1.3242	-1.3092
0.2382	2.0	150	0.2407	0.3257	-1.3197	0.9900	1.6454	-65.8341	-31.8276	-1.3243	-1.3092

Framework versions

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