Add description to card metadata

Crosslingual Optimized Metric for Evaluation of Translation (COMET) is an open-source framework used to train Machine Translation metrics that achieve high levels of correlation with different types of human judgments (HTER, DA's or MQM).
With the release of the framework the authors also released fully trained models that were used to compete in the WMT20 Metrics Shared Task achieving SOTA in that years competition.

See the [README.md] file at https://unbabel.github.io/COMET/html/models.html for more information.

README.md

@@ -1,6 +1,6 @@
 ---
 title: COMET
-emoji: 🤗 
+emoji: 🤗
 colorFrom: blue
 colorTo: red
 sdk: gradio
@@ -8,10 +8,22 @@ sdk_version: 3.0.2
 app_file: app.py
 pinned: false
 tags:
-- evaluate
-- metric
----
+  - evaluate
+  - metric
+description: >-
+  Crosslingual Optimized Metric for Evaluation of Translation (COMET) is an
+  open-source framework used to train Machine Translation metrics that achieve
+  high levels of correlation with different types of human judgments (HTER, DA's
+  or MQM).
+
+  With the release of the framework the authors also released fully trained
+  models that were used to compete in the WMT20 Metrics Shared Task achieving
+  SOTA in that years competition.
 
+
+  See the [README.md] file at https://unbabel.github.io/COMET/html/models.html
+  for more information.
+---
 # Metric Card for COMET
 
 ## Metric description