Minbyul
/

meditron-7b-dpo-full-wo-live_qa-ep3

Text Generation

alignment-handbook

Generated from Trainer

Inference Endpoints

text-generation-inference

Model card Files Files and versions Community

meditron-7b-dpo-full-wo-live_qa-ep3 / eval_results.json

Minbyul's picture

End of training

1e93d07 verified 2 months ago

raw history blame contribute delete

No virus

570 Bytes

	{
	"epoch": 1.0,
	"eval_logits/chosen": -0.8900327086448669,
	"eval_logits/rejected": -0.8877933025360107,
	"eval_logps/chosen": -986.1032104492188,
	"eval_logps/rejected": -1205.3543701171875,
	"eval_loss": 0.5355749130249023,
	"eval_rewards/accuracies": 0.692307710647583,
	"eval_rewards/chosen": -0.28710055351257324,
	"eval_rewards/margins": 0.4889349639415741,
	"eval_rewards/rejected": -0.7760356068611145,
	"eval_runtime": 47.8688,
	"eval_samples": 416,
	"eval_samples_per_second": 8.69,
	"eval_steps_per_second": 0.272
	}