Add BLOOMZ asset for OSACT4SubtaskB (#218)

Co-authored-by: Fahim Imaduddin Dalvi <[email protected]>

assets/ar/factuality_disinformation_harmful_content/hate_speech/OSACT4SubtaskB_BLOOMZ_ZeroShot.py

@@ -0,0 +1,39 @@
+import os
+
+from llmebench.datasets import OSACT4SubtaskBDataset
+from llmebench.models import PetalsModel
+from llmebench.tasks import HateSpeechTask
+
+
+def config():
+    return {
+        "dataset": OSACT4SubtaskBDataset,
+        "dataset_args": {},
+        "task": HateSpeechTask,
+        "task_args": {},
+        "model": PetalsModel,
+        "model_args": {
+            "class_labels": ["HS", "NOT_HS"],
+            "max_tries": 3,
+        },
+        "general_args": {
+            "data_path": "data/factuality_disinformation_harmful_content/hate_speech/OSACT2020-sharedTask-test-tweets-labels.txt"
+        },
+    }
+
+
+def prompt(input_sample):
+    return {
+        "prompt": 'Given the following Arabic tweet, label it as "HS" or "NOT_HS" based on the content of the tweet. Provide only label.\n\n'
+        + "sentence: "
+        + input_sample
+        + "label: \n"
+    }
+
+
+def post_process(response):
+    label = response["outputs"].strip()
+    label = label.replace("<s>", "")
+    label = label.replace("</s>", "")
+
+    return label