added more samples

lm_eval/tasks/super_glue/cb/t5-prompt.yaml

+group:
+  - super-glue-t5-prompt
+task: t5-prompt
+reference: "From Raffel et. al. 2019"
+dataset_path: super_glue
+dataset_name: cb
+training_split: train
+validation_split: validation
+doc_to_text: "cb hypothesis: {{hypothesis}} premise {{premise}}"
+doc_to_target: "{% set answer_choices = ['entailment', 'contradiction', 'neutral'] %}{{answer_choices[label]}}"
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+    ignore_case: true
+    ignore_punctuation: true

lm_eval/tasks/super_glue/copa/t5-prompt.yaml

+group:
+  - super-glue-t5-prompt
+task: t5-prompt
+reference: "From Raffel et. al. 2019"
+dataset_path: super_glue
+dataset_name: copa
+training_split: train
+validation_split: validation
+doc_to_text: "copa choice1: {{choice1}} choice2: {{choice2}} question: {{question}}"
+doc_to_target: "{% set answer_choices = ['False', 'True'] %}{{answer_choices[label]}}"
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+    ignore_case: true
+    ignore_punctuation: true

lm_eval/tasks/super_glue/record/t5-prompt.yaml

+group:
+  - super-glue-t5-prompt
+task: t5-prompt
+reference: "From Raffel et. al. 2019"
+dataset_path: super_glue
+dataset_name: record
+training_split: train
+validation_split: validation
+doc_to_text: "record query: {{query}} entities: {{entities}} passage: {{passage}}"
+doc_to_target: "{{answers}}"
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+    ignore_case: true
+    ignore_punctuation: true

lm_eval/tasks/super_glue/wsc/preprocess_wsc.py

+import re
+
+def doc_to_text(x):
+
+    def _mark_span(text, span_str, span_idx, mark):
+        pattern_tmpl = r'^((?:\S+\s){N})(W)'
+        pattern = re.sub(pattern_tmpl, 'N', str(span_idx))
+        pattern = re.sub(pattern, 'W', span_str)
+        return re.sub(text, pattern, r'\1{0} \2 {0}'.format(mark))
+
+    text = x['text']
+    text = _mark_span(text, x['span1_text'], x['span1_index'], '*')
+    # Compensate for 2 added "words" added in previous step.
+    span2_index = x['span2_index'] + 2 * (x['span1_index'] < x['span2_index'])
+    text = _mark_span(text, x['span2_text'], span2_index, '#')
+
+    return text
\ No newline at end of file

lm_eval/tasks/super_glue/wsc/t5-prompt.yaml

+group:
+  - super-glue-t5-prompt
+task: t5-prompt
+reference: "From Raffel et. al. 2019"
+dataset_path: super_glue
+dataset_name: wsc
+training_split: train
+validation_split: validation
+doc_to_text: !function "preprocess_wsc.doc_to_text"
+doc_to_target: "{{candidates[label]}}"
+metric_list:
+  - metric: exact_match
+    aggregation: mean
+    higher_is_better: true
+    ignore_case: true
+    ignore_punctuation: true