tsavage68/MedQA_L3_300steps_1e6rate_03beta_CSFTDPO

MedQA_L3_1000steps_1e6rate_03beta_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: 0.4888
• Rewards/chosen: 3.1508
• Rewards/rejected: 1.3776
• Rewards/accuracies: 0.7868
• Rewards/margins: 1.7732
• Logps/rejected: -29.2628
• Logps/chosen: -20.8258
• Logits/rejected: -0.8475
• Logits/chosen: -0.8455

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: 300

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.685	0.0489	50	0.6334	-0.7936	-0.9359	0.7363	0.1423	-36.9746	-33.9739	-0.7278	-0.7271
0.4052	0.0977	100	0.6106	3.7995	2.4858	0.6945	1.3137	-25.5688	-18.6634	-0.7922	-0.7909
0.6421	0.1466	150	0.5225	2.5850	1.3506	0.7538	1.2344	-29.3529	-22.7119	-0.8369	-0.8356
0.3501	0.1954	200	0.5243	2.6639	0.8481	0.7626	1.8159	-31.0279	-22.4487	-0.8442	-0.8422
0.3618	0.2443	250	0.4899	3.1411	1.3754	0.7802	1.7657	-29.2702	-20.8582	-0.8474	-0.8454
0.3181	0.2931	300	0.4888	3.1508	1.3776	0.7868	1.7732	-29.2628	-20.8258	-0.8475	-0.8455

Framework versions

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