made more generic, aligned with other tasks

examples/evaluate_ict_zeroshot_nq.sh

@@ -12,7 +12,9 @@ CHECKPOINT_PATH=<Specify path of pretrained ICT model>
 
 QA_FILE=<Path of the natural question test dataset>
 
-python tasks/orqa/evaluate_orqa.py \
+python tasks/main.py \
+    --task ICT-ZEROSHOT-NQ \
+    --tokenizer-type BertWordPieceLowerCase \
     --num-layers 12 \
     --hidden-size 768 \
     --num-attention-heads 12 \

tasks/main.py

@@ -62,6 +62,8 @@ if __name__ == '__main__':
         from glue.finetune import main
     elif args.task in ['LAMBADA', 'WIKITEXT103']:
         from zeroshot_gpt.evaluate import main
+    elif args.task in ['ICT-ZEROSHOT-NQ']:
+        from orqa.evaluate_orqa import main
     else:
         raise NotImplementedError('Task {} is not implemented.'.format(
             args.task))

tasks/orqa/evaluate_orqa.py

@@ -18,9 +18,6 @@
 import os
 import sys
 
-sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__),
-    os.path.join(os.path.pardir, os.path.pardir))))
-
 from megatron import get_args
 from megatron.initialize import initialize_megatron
 
@@ -30,8 +27,7 @@ def main():
     """
     Main program
     """
-    initialize_megatron(extra_args_provider=None,
-                        args_defaults={'tokenizer_type': 'BertWordPieceLowerCase'})
+
     args = get_args()
 
     # Set up the model and evaluator