From c511308764ffed4d8fad30e2cd7e25fbb0d947b3 Mon Sep 17 00:00:00 2001 From: Francesco Bertolotti Date: Tue, 18 Nov 2025 16:56:35 +0100 Subject: [PATCH] batched metric was not aggregated properly Co-authored-by: Francesco Bertolotti --- src/lighteval/metrics/__init__.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/src/lighteval/metrics/__init__.py b/src/lighteval/metrics/__init__.py index da0cff862..d61b13764 100644 --- a/src/lighteval/metrics/__init__.py +++ b/src/lighteval/metrics/__init__.py @@ -45,8 +45,8 @@ def apply_metric(responses: list[ModelResponse], docs: list[Doc], metrics: list[ output = {} # Add batched metric results for this sample - for metric_outputs in batched_outputs: - output.update(metric_outputs[i]) + for metric, metric_outputs in zip(batched_metrics, batched_outputs): + output.update({metric.metric_name: metric_outputs[metric.metric_name][i]}) # Add non-batched metric results for this sample for metric in non_batched_metrics: