Test

README.md

@@ -9,17 +9,6 @@ metrics:
 model-index:
 - name: autoevaluate-binary-classification
   results:
-  - task:
-      name: Text Classification
-      type: text-classification
-    dataset:
-      name: glue
-      type: glue
-      args: sst2
-    metrics:
-    - name: Accuracy
-      type: accuracy
-      value: 0.8967889908256881
   - task:
       type: text-classification
       name: Text Classification
@@ -29,33 +18,39 @@ model-index:
       config: sst2
       split: validation
     metrics:
-    - name: Accuracy
-      type: accuracy
+    - type: accuracy
+      value: 0.8967889908256881
+      name: Accuracy
+      verified: true
+      verifyToken: 1234
+    - type: accuracy
       value: 0.8967889908256881
+      name: Accuracy
       verified: true
-    - name: Precision
-      type: precision
+      verifyToken: 1234
+    - type: precision
       value: 0.8898678414096917
+      name: Precision
       verified: true
-    - name: Recall
-      type: recall
+    - type: recall
       value: 0.9099099099099099
+      name: Recall
       verified: true
-    - name: AUC
-      type: auc
+    - type: auc
       value: 0.967247621453229
+      name: AUC
       verified: true
-    - name: F1
-      type: f1
+    - type: f1
       value: 0.8997772828507795
+      name: F1
       verified: true
-    - name: loss
-      type: loss
+    - type: loss
       value: 0.30091655254364014
+      name: loss
       verified: true
-    - name: matthews_correlation
-      type: matthews_correlation
+    - type: matthews_correlation
       value: 0.793630584795814
+      name: matthews_correlation
       verified: true
 ---