Merge pull request huggingface#12 from stevezheng23/dev/zheng/quac

Dev/zheng/quac

examples/run_quac_kd.py

@@ -361,6 +361,7 @@ def predict(args, model, tokenizer, prefix=""):
         result = result_lookup[feature.unique_id]
         feature.start_targets = _compute_softmax(result.kd_start_logits)
         feature.end_targets = _compute_softmax(result.kd_end_logits)
+        updated_features.append(feature)
 
     torch.save(updated_features, updated_features_file)
 
@@ -471,6 +472,8 @@ def main():
                         help="Whether to run training.")
     parser.add_argument("--do_eval", action='store_true',
                         help="Whether to run eval on the dev set.")
+    parser.add_argument("--do_predict", action='store_true',
+                        help="Whether to run predict on the dev set.")
     parser.add_argument("--evaluate_during_training", action='store_true',
                         help="Rul evaluation during training at each logging step.")
     parser.add_argument("--do_lower_case", action='store_true',

examples/utils_quac.py

@@ -948,9 +948,9 @@ def write_predictions_v2(all_examples, all_features, all_results, n_best_size,
 
         if null_score_threshold is not None:
             if nbest_json[0]["text"] == "CANNOTANSWER" and nbest_json[0]["probability"] > null_score_threshold:
-                all_predictions[qas_id] = "CANNOTANSWER"
+                all_predictions[example.qas_id] = "CANNOTANSWER"
             else:
-                all_predictions[qas_id] = nbest_json[0]["text"] if nbest_json[0]["text"] != "CANNOTANSWER" else nbest_json[1]["text"]
+                all_predictions[example.qas_id] = nbest_json[0]["text"] if nbest_json[0]["text"] != "CANNOTANSWER" else nbest_json[1]["text"]
         else:
             if not version_2_with_negative:
                 all_predictions[example.qas_id] = nbest_json[0]["text"]