MedQA_L3_350steps_1e7rate_01beta_CSFTDPO

This model is a fine-tuned version of tsavage68/MedQA_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6777
• Rewards/chosen: 0.1095
• Rewards/rejected: 0.0772
• Rewards/accuracies: 0.7055
• Rewards/margins: 0.0324
• Logps/rejected: -33.0833
• Logps/chosen: -30.2335
• Logits/rejected: -0.7312
• Logits/chosen: -0.7305

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: 350

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6932	0.0489	50	0.6927	-0.0017	-0.0025	0.5297	0.0008	-33.8801	-31.3453	-0.7320	-0.7313
0.691	0.0977	100	0.6894	0.0852	0.0776	0.6505	0.0076	-33.0791	-30.4769	-0.7328	-0.7321
0.6769	0.1466	150	0.6822	0.1412	0.1183	0.6857	0.0228	-32.6716	-29.9169	-0.7316	-0.7309
0.6718	0.1954	200	0.6794	0.0847	0.0559	0.7011	0.0288	-33.2958	-30.4811	-0.7309	-0.7302
0.6835	0.2443	250	0.6781	0.1060	0.0745	0.6791	0.0316	-33.1100	-30.2681	-0.7308	-0.7300
0.6749	0.2931	300	0.6777	0.1081	0.0756	0.7055	0.0325	-33.0987	-30.2473	-0.7318	-0.7311
0.6792	0.3420	350	0.6777	0.1095	0.0772	0.7055	0.0324	-33.0833	-30.2335	-0.7312	-0.7305

Framework versions

• Transformers 4.41.1
• Pytorch 2.0.0+cu117
• Datasets 2.19.1
• Tokenizers 0.19.1