fix to t5 adapter and ia3 evals due to predict_step dictionary key ch…

…anges (#5193) Signed-off-by: arendu <adithya.r@gmail.com> Signed-off-by: arendu <adithya.r@gmail.com>
NVIDIA · Oct 18, 2022 · acb5073 · acb5073
1 parent 8242d72
commit acb5073
Show file tree

Hide file tree

Showing 3 changed files with 10 additions and 6 deletions.
diff --git a/Jenkinsfile b/Jenkinsfile
@@ -701,7 +701,8 @@ pipeline {
                 pipeline_model_parallel_size=2 \
                 data.global_batch_size=2 \
                 data.micro_batch_size=2 \
-                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl']"
+                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl'] \
+                pred_file_path='examples/adapter_tuning/test_tp1_pp2/preds.txt'"
             sh "rm -rf examples/adapter_tuning/test_tp1_pp2.nemo"
             sh "rm -rf examples/adapter_tuning/test_tp1_pp2"
           }
@@ -742,7 +743,8 @@ pipeline {
                 tensor_model_parallel_size=2 \
                 data.global_batch_size=2 \
                 data.micro_batch_size=2 \
-                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl']"
+                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl'] \
+                pred_file_path='examples/adapter_tuning/test_tp2_pp1/preds.txt'"
             sh "rm -rf examples/adapter_tuning/test_tp2_pp1.nemo"
             sh "rm -rf examples/adapter_tuning/test_tp2_pp1"
           }
@@ -785,7 +787,8 @@ pipeline {
                 pipeline_model_parallel_size=2 \
                 data.global_batch_size=2 \
                 data.micro_batch_size=2 \
-                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl']"
+                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl'] \
+                pred_file_path='examples/ia3_tuning/test_tp1_pp2/preds.txt'"
             sh "rm -rf examples/ia3_tuning/test_tp1_pp2.nemo"
             sh "rm -rf examples/ia3_tuning/test_tp1_pp2"
           }
@@ -826,7 +829,8 @@ pipeline {
                 tensor_model_parallel_size=2 \
                 data.global_batch_size=2 \
                 data.micro_batch_size=2 \
-                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl']"
+                data.test_ds=['/home/TestData/nlp/prompt_learning/rte_CI_test.jsonl'] \
+                pred_file_path='examples/ia3_tuning/test_tp2_pp1/preds.txt'"
             sh "rm -rf examples/ia3_tuning/test_tp2_pp1.nemo"
             sh "rm -rf examples/ia3_tuning/test_tp2_pp1"
           }

diff --git a/examples/nlp/language_modeling/tuning/megatron_t5_adapter_eval.py b/examples/nlp/language_modeling/tuning/megatron_t5_adapter_eval.py
@@ -129,7 +129,7 @@ def dummy():
     if cfg.pred_file_path is not None:
         with open(cfg.pred_file_path, "w", encoding="utf-8") as f:
             for batch in response:
-                for inp, pred in zip(batch['enc_input'], batch['predicted_token_ids']):
+                for inp, pred in zip(batch['input_text'], batch['preds_text']):
                     inp = ' '.join(inp.split('\n'))
                     pred = ' '.join(pred.split('\n'))
                     f.write(f'{inp} {pred}\n')

diff --git a/examples/nlp/language_modeling/tuning/megatron_t5_ia3_eval.py b/examples/nlp/language_modeling/tuning/megatron_t5_ia3_eval.py
@@ -129,7 +129,7 @@ def dummy():
     if cfg.pred_file_path is not None:
         with open(cfg.pred_file_path, "w", encoding="utf-8") as f:
             for batch in response:
-                for inp, pred in zip(batch['enc_input'], batch['predicted_token_ids']):
+                for inp, pred in zip(batch['input_text'], batch['preds_text']):
                     inp = ' '.join(inp.split('\n'))
                     pred = ' '.join(pred.split('\n'))
                     f.write(f'{inp} {pred}\n')