diff --git a/.codex b/.codex
new file mode 100644
index 00000000..e69de29b
diff --git a/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen.py b/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen.py
index 41d7dd63..1d6d4b5e 100644
--- a/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen.py
+++ b/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen.py
@@ -3,7 +3,7 @@
 from ais_bench.benchmark.openicl.icl_inferencer import GenInferencer
 from ais_bench.benchmark.datasets import TEXTVQADataset, TEXTEvaluator, TEXTEvaluatorForVita
 
-
+PROMPT = "Answer the question using a single word or phrase."
 textvqa_reader_cfg = dict(
     input_columns=['question', 'image'],
     output_column='answer'
@@ -16,7 +16,7 @@
         template=dict(
             round=[
                 dict(role="HUMAN", prompt_mm={
-                    "text": {"type": "text", "text": "{question} Answer the question using a single word or phrase."},
+                    "text": {"type": "text", "text": f"{{question}} {PROMPT}"},
                     "image": {"type": "image_url", "image_url": {"url": "file://{image}"}},
                 })
             ]
diff --git a/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen_base64.py b/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen_base64.py
index 1e567370..6ed3a76c 100644
--- a/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen_base64.py
+++ b/ais_bench/benchmark/configs/datasets/textvqa/textvqa_gen_base64.py
@@ -3,6 +3,7 @@
 from ais_bench.benchmark.openicl.icl_inferencer import GenInferencer
 from ais_bench.benchmark.datasets import TEXTVQADataset, TEXTEvaluator
 
+PROMPT = "Answer the question using a single word or phrase."
 
 textvqa_reader_cfg = dict(
     input_columns=['question', 'image'],
@@ -16,7 +17,7 @@
         template=dict(
             round=[
                 dict(role="HUMAN", prompt_mm={
-                    "text": {"type": "text", "text": "{question} Answer the question using a single word or phrase."},
+                    "text": {"type": "text", "text": f"{{question}} {PROMPT}"},
                     "image": {"type": "image_url", "image_url": {"url": "data:image/jpeg;base64,{image}"}},
                 })
             ]