Update ADI BLOOMZ asset (#181)

Updated prompt and post-processing to improve performance.

assets/ar/sequence_tagging_and_information_extraction/dialect_identification/ADI_BLOOMZ_ZeroShot.py

@@ -17,37 +17,12 @@ def config():
         "dataset": ADIDataset,
         "task": DialectIDTask,
         "model": PetalsModel,
-        "model_args": {
-            "class_labels": [
-                "EGY",
-                "IRA",
-                "JOR",
-                "KSA",
-                "KUW",
-                "LEB",
-                "LIB",
-                "MOR",
-                "MSA",
-                "PAL",
-                "QAT",
-                "SUD",
-                "SYR",
-                "UAE",
-                "YEM",
-            ],
-            "max_tries": 3,
-        },
     }
 
 
 def prompt(input_sample):
-    arr = input_sample.split()
-    if len(arr) > 500:
-        input_sample = arr[:500]
-
     prompt_string = (
-        f'Classify the following "text" into one of the following categories: "EGY", "IRA", "JOR", "KSA", "KUW", "LEB", "LIB", "MOR", "MSA", "PAL", "QAT", "SUD", "SYR", "UAE", "YEM"\n'
-        f"Please provide only the label.\n\n"
+        f'Identify the dialect of the following Arabic "text" given the following possible dialects: "Egyptian", "Iraqi", "Jordanian", "Saudi", "Kuwaiti", "Lebanese", "Libyan", "Moroccan", "modern standard Arabic", "Palestinian", "Qatari", "Sudanese", "Syrian", "Emirati", "Yemeni"\n'
         f"text: {input_sample}\n"
         f"label: \n"
     )
@@ -58,45 +33,36 @@ def prompt(input_sample):
 
 
 def post_process(response):
+    count_label_map = {
+        "Egyptian": "EGY",
+        "Iraqi": "IRA",
+        "Jordanian": "JOR",
+        "Saudi": "KSA",
+        "Kuwaiti": "KUW",
+        "Lebanese": "LEB",
+        "Libyan": "LIB",
+        "Moroccan": "MOR",
+        "modern standard Arabic": "MSA",
+        "Modern standard Arabic": "MSA",
+        "Modern Standard Arabic": "MSA",
+        "Palestinian": "PAL",
+        "Qatari": "QAT",
+        "Sudanese": "SUD",
+        "Syrian": "SYR",
+        "Emirati": "UAE",
+        "Yemeni": "YEM",
+        "Yemen": "YEM",
+    }
+
     label = response["outputs"].strip()
     label = label.replace("<s>", "")
     label = label.replace("</s>", "")
-    label = label.lower()
-
-    # label_list = config()["model_args"]["class_labels"]
-    # label_list = [lab.lower() for lab in label_list]
-    #
-    # if "label: " in label:
-    #     label_fixed = label.replace("label: ", "").lower()
-    # elif label.lower() in label_list:
-    #     label_fixed = label.lower()
-    # else:
-    #     label_fixed = None
-    label_list = config()["model_args"]["class_labels"]
-    label_list = [dialect.lower() for dialect in label_list]
-
-    label = label.replace("label:", "").strip()
+    label = label.replace("Dialect: ", "").replace("dialect: ", "")
+    label = label.replace("label: ", "")
+    label = label.strip()
 
-    if label in label_list:
-        label_fixed = label
-    elif "\n msa" in label:
-        label_fixed = "msa"
-    elif "\n ksa" in label:
-        label_fixed = "ksa"
-    elif "\n pal" in label:
-        label_fixed = "pal"
-    elif "\n egy" in label:
-        label_fixed = "egy"
-    elif "\n yem" in label:
-        label_fixed = "yem"
-    elif "\n syr" in label:
-        label_fixed = "syr"
-    elif "\n jor" in label:
-        label_fixed = "jor"
-    elif "\n ira" in label:
-        label_fixed = "ira"
-    elif "\n kuw" in label:
-        label_fixed = "kuw"
+    if label in count_label_map:
+        label_fixed = count_label_map[label].lower()
     else:
         label_fixed = None