chore(dataset): remove prompt

mvl-sib200.py

@@ -68,20 +68,6 @@ CATEGORIES: List[str] = [
     "travel",
 ]
 
-# Prompts for classification tasks
-IMG2SENT_PROMPT: str = (
-    "Which option best matches the topic of the reference image(s)? "
-    'The available topics are "entertainment", "geograpy", "health", '
-    '"politics", "science and technology", "sports", and "travel". '
-    "Choose one from A, B, C, D and only output a single letter."
-)
-SENT2IMG_PROMPT: str = (
-    "Which option best matches the topic of the reference sentence(s)? "
-    'The available topics are "entertainment", "geograpy", "health", '
-    '"politics", "science and technology", "sports", and "travel". '
-    "Choose one from A, B, C, D and only output a single letter."
-)
-
 # URLs for downloading SIB .tsv data and images.
 _SIB_URL: str = "https://huggingface.co/datasets/wuenlp/mvl-sib200/resolve/main/data/sib200/{lang}/{split}.tsv"
 _IMG_URL: str = "https://huggingface.co/datasets/wuenlp/mvl-sib200/resolve/main/data/images/sib200/{category}_{no}.jpg"
@@ -277,7 +263,7 @@ def replicate_and_negatives(
 
     for cat, group_df in grouped:
         g_size = len(group_df)
-        
+
         # The preallocated arrays for negative and positive columns will be filled for each row individually, i.e., sampling of negative categories and samples will be done per row
         # Prepare arrays for final negative columns
         neg_id_cols = [np.empty(g_size, dtype=object) for _ in range(num_negatives)]
@@ -472,7 +458,6 @@ class MVLSIB(datasets.GeneratorBasedBuilder):
     The dataset is structured such that each row includes:
         - A set of reference items (images or sentences, depending on the task).
         - A set of 4 possible answers (1 positive, 3 negative).
-        - A prompt instructing the user to pick the correct match.
         - A label indicating which of the 4 answers is correct.
     """
 
@@ -494,7 +479,6 @@ class MVLSIB(datasets.GeneratorBasedBuilder):
           - label (int specifying which of the sentences is correct)
           - id (an integer ID)
           - index_id (the original row ID from the SIB .tsv)
-          - prompt (str, the classification prompt)
 
         Each example row in 'sent2img' includes:
           - sentences (list of str, the positive reference sentences)
@@ -503,7 +487,6 @@ class MVLSIB(datasets.GeneratorBasedBuilder):
           - label (int specifying which of the images is correct)
           - id (an integer ID)
           - index_id (the original row ID from the SIB .tsv)
-          - prompt (str, the classification prompt)
 
         Returns
         -------
@@ -518,7 +501,6 @@ class MVLSIB(datasets.GeneratorBasedBuilder):
                 "images": Sequence(Value("string")),
                 "sentences": Sequence(Value("string")),
                 "categories": Sequence(Value("string")),
-                "prompt": Value("string"),
                 "label": Value("int8"),
                 "id": Value("int64"),
                 "index_id": Value("int64"),
@@ -532,7 +514,6 @@ class MVLSIB(datasets.GeneratorBasedBuilder):
                 "label": Value("int8"),
                 "id": Value("int64"),
                 "index_id": Value("int64"),
-                "prompt": Value("string"),
             }
         )
 
@@ -673,7 +654,6 @@ class MVLSIB(datasets.GeneratorBasedBuilder):
                     i,
                     {
                         "id": i,
-                        "prompt": IMG2SENT_PROMPT,
                         "index_id": row["index_id"],
                         "images": row_images,
                         "categories": categories_shuffled,
@@ -717,7 +697,6 @@ class MVLSIB(datasets.GeneratorBasedBuilder):
                     i,
                     {
                         "id": i,
-                        "prompt": SENT2IMG_PROMPT,
                         "index_id": row["index_id"],
                         "images": row_images,
                         "categories": categories_shuffled,