MedQA_L3_250steps_1e6rate_01beat_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: 0.4710
• Rewards/chosen: -0.7540
• Rewards/rejected: -1.6509
• Rewards/accuracies: 0.7758
• Rewards/margins: 0.8969
• Logps/rejected: -37.8254
• Logps/chosen: -25.7624
• Logits/rejected: -1.1604
• Logits/chosen: -1.1585

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: 250

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.695	0.0489	50	0.6713	0.0342	-0.0142	0.6615	0.0484	-21.4583	-17.8807	-0.9400	-0.9395
0.6187	0.0977	100	0.5915	-0.1174	-0.4200	0.7121	0.3027	-25.5168	-19.3963	-1.0412	-1.0403
0.559	0.1466	150	0.5116	-0.4993	-1.1517	0.7429	0.6524	-32.8335	-23.2153	-1.1115	-1.1101
0.4654	0.1954	200	0.4732	-0.7696	-1.6630	0.7780	0.8934	-37.9465	-25.9187	-1.1618	-1.1598
0.4766	0.2443	250	0.4710	-0.7540	-1.6509	0.7758	0.8969	-37.8254	-25.7624	-1.1604	-1.1585

Framework versions

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