MedQA_L3_300steps_1e7rate_05beta_CSFTDPO

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6479
• Rewards/chosen: 0.2870
• Rewards/rejected: 0.1538
• Rewards/accuracies: 0.6374
• Rewards/margins: 0.1332
• Logps/rejected: -21.0089
• Logps/chosen: -17.6487
• Logits/rejected: -0.9327
• Logits/chosen: -0.9321

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: 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
0.7034	0.0489	50	0.6908	0.0092	0.0030	0.5187	0.0061	-21.3104	-18.2043	-0.9262	-0.9257
0.6841	0.0977	100	0.6705	0.1777	0.1221	0.6088	0.0556	-21.0723	-17.8673	-0.9278	-0.9273
0.6636	0.1466	150	0.6536	0.2698	0.1543	0.6505	0.1155	-21.0080	-17.6830	-0.9307	-0.9302
0.6483	0.1954	200	0.6488	0.2862	0.1570	0.6330	0.1291	-21.0025	-17.6503	-0.9322	-0.9317
0.683	0.2443	250	0.6472	0.2913	0.1569	0.6396	0.1344	-21.0027	-17.6400	-0.9325	-0.9320
0.6269	0.2931	300	0.6479	0.2870	0.1538	0.6374	0.1332	-21.0089	-17.6487	-0.9327	-0.9321

Framework versions

• Transformers 4.41.0
• Pytorch 2.0.0+cu117
• Datasets 2.19.1
• Tokenizers 0.19.1