Update README.md

README.md

@@ -642,6 +642,13 @@ The datasets are divided into subsets based on context lengths: 4k, 8k, 16k, 32k
 - **ruGSM100**: Solve math problems using Chain-of-Thought reasoning. Created by translating the [GSM100](https://huggingface.co/datasets/L4NLP/LEval/viewer/gsm100) dataset from L-Eval.
 
 
+## Metrics
+
+The benchmark employs a range of evaluation metrics to ensure comprehensive assessment across different tasks and context lengths. These metrics are designed to measure the models' capabilities in handling long-context understanding, reasoning, and extraction of relevant information from extensive texts. Here are the primary metrics used in this benchmark:
+
+- **Exact Match (EM)**: This metric is used to evaluate the accuracy of the model's responses by comparing the predicted answers to the ground truth. It is particularly effective for tasks where precise matching of responses is critical, such as question answering and retrieval tasks.
+
+- **Overall Score**: For each model, the overall score is calculated by averaging the scores obtained across various tasks and context lengths. This provides a holistic view of the model's performance and its ability to generalize across different types of long-context tasks.
 
 ## Usage