Minbyul
/

llama2-7b-dpo-full-wo-kqa_silver_wogold-ep3

Text Generation

alignment-handbook

Generated from Trainer

Inference Endpoints

text-generation-inference

Model card Files Files and versions Community

llama2-7b-dpo-full-wo-kqa_silver_wogold-ep3 / eval_results.json

Minbyul's picture

End of training

86bce7f verified 2 months ago

raw history blame contribute delete

No virus

574 Bytes

	{
	"epoch": 1.0,
	"eval_logits/chosen": 0.47494596242904663,
	"eval_logits/rejected": -0.5924796462059021,
	"eval_logps/chosen": -408.63677978515625,
	"eval_logps/rejected": -1010.3619384765625,
	"eval_loss": 0.6588297486305237,
	"eval_rewards/accuracies": 0.7911764979362488,
	"eval_rewards/chosen": 0.047554861754179,
	"eval_rewards/margins": 0.07667416334152222,
	"eval_rewards/rejected": -0.029119295999407768,
	"eval_runtime": 302.3402,
	"eval_samples": 2712,
	"eval_samples_per_second": 8.97,
	"eval_steps_per_second": 0.281
	}