diff --git a/scripts/base_eval.py b/scripts/base_eval.py index 6e61a95..8b83910 100644 --- a/scripts/base_eval.py +++ b/scripts/base_eval.py @@ -177,7 +177,7 @@ def main(): # Log to report from nanochat.report import get_report - get_report().log(section="Base model evaluation", data=[ + get_report(exp_name=model_tag).log(section="Base model evaluation", data=[ { "Model": model_name, "CORE metric": core_metric, diff --git a/scripts/chat_eval.py b/scripts/chat_eval.py index 03d34c3..e91332f 100644 --- a/scripts/chat_eval.py +++ b/scripts/chat_eval.py @@ -245,7 +245,7 @@ if __name__ == "__main__": centered_mean += centered_acc chatcore_metric = centered_mean / len(results) chatcore_metric_dict = {"ChatCORE metric": chatcore_metric} - get_report().log(section="Chat evaluation " + args.source, data=[ + get_report(exp_name=args.model_tag).log(section="Chat evaluation " + args.source, data=[ vars(args), # CLI args results, chatcore_metric_dict,