update on afrimgsm

lm_eval/tasks/afrimgsm/gen_yaml.sh

+#!/bin/bash
+
+python utils.py --overwrite --output-dir direct --mode direct
+# python utils.py --overwrite --output-dir en_cot --mode en-cot
+# python utils.py --overwrite --output-dir native_cot --mode native-cot

lm_eval/tasks/afrimgsm/gpt_direct.sh

-#!/bin/bash
-
-models=(
-  "gpt-3.5-turbo"
-  "gpt-4-0125-preview"
-)
-task=afrimgsm_direct_amh,afrimgsm_direct_eng,afrimgsm_direct_ewe,afrimgsm_direct_fra,afrimgsm_direct_hau,afrimgsm_direct_ibo,afrimgsm_direct_kin,afrimgsm_direct_lin,afrimgsm_direct_lug,afrimgsm_direct_orm,afrimgsm_direct_sna,afrimgsm_direct_sot,afrimgsm_direct_swa,afrimgsm_direct_twi,afrimgsm_direct_wol,afrimgsm_direct_xho,afrimgsm_direct_yor,afrimgsm_direct_zul
-
-for model in "${models[@]}"
-do
-  echo "Evaluating model: $model"
-  for fewshot in 0 2 4 6 8
-  do
-    export OUTPUT_DIR=results/$fewshot
-
-    mkdir -p "$OUTPUT_DIR"
-
-    lm_eval --model openai-chat-completions \
-            --model_args model="${model}" \
-            --tasks $task \
-            --output_path "$OUTPUT_DIR" \
-            --num_fewshot $fewshot \
-            --verbosity DEBUG
-  done
-done
\ No newline at end of file

lm_eval/tasks/afrimgsm/utils.py

 import argparse
-import yaml
 
+import yaml
 languages = ['eng', 'amh', 'ibo', 'fra', 'sna', 'lin', 'wol', 'ewe', 'lug', 'xho', 'kin', 'twi', 'zul', 'orm', 'yor', 'hau', 'sot', 'swa']
 
-configs = {
-    "QUESTION": "Question:",
-    "ANSWER": "Step-by-Step Answer:",
-    "DIRECT": "Answer:",
-    "REGEX": "The answer is (\\-?[0-9\\.\\,]+)"}
+LANGUAGES = {}
+
+for lang in languages:
+    LANGUAGES[lang] = {  # English
+        "QUESTION": "Question:",
+        "ANSWER": "Step-by-Step Answer:",
+        "DIRECT": "Answer:",
+        "REGEX": "The answer is (\\-?[0-9\\.\\,]+)"} 
+
+
+def add_regex_pattern(regex_pattern):
+    if regex_pattern is None:
+        return {}
+    return {
+        "filter_list": [
+            {
+                "name": "strict-match",
+                "filter": [
+                    {
+                        "function": "regex",
+                        "regex_pattern": f"""{regex_pattern}""",
+                    },
+                    {
+                        "function": "take_first",
+                    },
+                ],
+            },
+            {
+                "name": "flexible-extract",
+                "filter": [
+                    {
+                        "function": "regex",
+                        "regex_pattern": """(-?[$0-9.,]{2,})|(-?[0-9]+)""",
+                        "group_select": -1,
+                    },
+                    {
+                        "function": "take_first",
+                    },
+                ],
+            },
+        ],
+    }
 
 
 def gen_lang_yamls(output_dir: str, overwrite: bool, mode: str) -> None:
@@ -18,19 +55,31 @@ def gen_lang_yamls(output_dir: str, overwrite: bool, mode: str) -> None:
     :param overwrite: Whether to overwrite files if they already exist.
     """
     err = []
-    for lang in languages:
+    for lang in LANGUAGES.keys():
         try:
+            QUESTION = LANGUAGES[lang]["QUESTION"]
+
+            yaml_template = "cot_yaml"
+            filter_list = {}
+            DELIMITER = None
             if mode == "direct":
+                ANSWER = LANGUAGES[lang]["DIRECT"]
+                REGEX = None
                 task_name = f"afrimgsm_direct_{lang}"
-                yaml_template = "afrimgsm_common_yaml"
+                yaml_template = "direct_yaml"
             elif mode == "native-cot":
+                ANSWER = LANGUAGES[lang]["ANSWER"]
+                REGEX = LANGUAGES[lang]["REGEX"]
                 task_name = f"afrimgsm_native_cot_{lang}"
-                yaml_template = "afrimgsm_common_yaml"
+                filter_list = add_regex_pattern(REGEX)
+                DELIMITER = "" if lang in ["zh", "ja"] else None
             elif mode == "en-cot":
+                ANSWER = LANGUAGES["en"]["ANSWER"]
+                REGEX = LANGUAGES["en"]["REGEX"]
                 task_name = f"afrimgsm_en_cot_{lang}"
-                yaml_template = "afrimgsm_common_yaml"
 
             file_name = f"{task_name}.yaml"
+            ANSWER_TO_SKIP = len(LANGUAGES[lang]["ANSWER"]) + 1
             with open(
                 f"{output_dir}/{file_name}", "w" if overwrite else "x", encoding="utf8"
             ) as f:
@@ -39,7 +88,23 @@ def gen_lang_yamls(output_dir: str, overwrite: bool, mode: str) -> None:
                     {
                         "include": yaml_template,
                         "dataset_name": lang,
-                        "task": f"{task_name}"
+                        "task": f"{task_name}",
+                        "doc_to_text": f"""{{% if answer is not none %}}"""
+                        f"""{{{{question+"\\n{ANSWER}"}}}}"""
+                        f"""{{% else %}}"""
+                        f"""{{{{"{QUESTION} "+question+"\\n{ANSWER}"}}}}"""
+                        f"""{{% endif %}}""",
+                        "doc_to_target": f"""{{% if answer is not none %}}"""
+                        f"""{{{{answer[{ANSWER_TO_SKIP}:]}}}}"""
+                        f"""{{% else %}}"""
+                        f"""{{{{answer_number|string}}}}"""
+                        f"""{{% endif %}}""",
+                        **filter_list,
+                        "generation_kwargs": {
+                            "until": [QUESTION, "</s>", "<|im_end|>"],
+                            "do_sample": False,
+                        },
+                        **({"target_delimiter": DELIMITER} if DELIMITER else {}),
                     },
                     f,
                     allow_unicode=True,
@@ -60,16 +125,16 @@ def main() -> None:
     parser = argparse.ArgumentParser()
     parser.add_argument(
         "--overwrite",
-        default=True,
+        default=False,
         action="store_true",
         help="Overwrite files if they already exist",
     )
     parser.add_argument(
-        "--output-dir", default="./direct", help="Directory to write yaml files to"
+        "--output-dir", default=".", help="Directory to write yaml files to"
     )
     parser.add_argument(
         "--mode",
-        default="direct",
+        default="native-cot",
         choices=["direct", "native-cot", "en-cot"],
         help="Mode of chain-of-thought",
     )