Translate maime prompt templates to target languages (fi, da)

The instruction prompts were in English for all language variants.
Now Finnish tasks get Finnish prompts and Danish tasks get Danish prompts.
The boxed answer format tokens are kept as-is for scorer compatibility.

Author: Daniel Zautner

src/lighteval/tasks/multilingual/tasks/maime.py

@@ -33,39 +33,53 @@
 from lighteval.tasks.requests import Doc
 
 
-# Prompt template adapted from AIME task
-# Note: Uses English instructions for consistency with AIME
-MATH_PROMPT_TEMPLATE = dedent("""
-Solve the following math problem efficiently and clearly.  
-The last line of your response should be of the following format: 
-'Therefore, the final answer is: $\\boxed{{ANSWER}}$. I hope it is correct' 
-(without quotes) where ANSWER is just the final number or expression 
-that solves the problem. Think step by step before answering.
+# Per-language prompt templates
+MATH_PROMPT_TEMPLATES = {
+    "fi": dedent("""
+Ratkaise seuraava matemaattinen tehtävä tehokkaasti ja selkeästi.
+Vastauksesi viimeisen rivin tulee olla seuraavassa muodossa:
+'Näin ollen lopullinen vastaus on: $\\boxed{{ANSWER}}$. Toivottavasti se on oikein'
+(ilman lainausmerkkejä), jossa ANSWER on pelkästään lopullinen luku tai lauseke,
+joka ratkaisee tehtävän. Ajattele vaihe vaiheelta ennen vastaamista.
 
 {prompt}
-""")
+"""),
+    "da": dedent("""
+Løs følgende matematiske problem effektivt og tydeligt.
+Den sidste linje i dit svar skal være i følgende format:
+'Derfor er det endelige svar: $\\boxed{{ANSWER}}$. Jeg håber, det er korrekt'
+(uden anførselstegn), hvor ANSWER kun er det endelige tal eller udtryk,
+der løser problemet. Tænk trin for trin, før du svarer.
 
+{prompt}
+"""),
+}
 
 
 def record_to_sample(record):
     return Sample(input=record["question"], target=record["solution"])
 
 
-def maime_prompt(line, task_name: str = None):
-    return Doc(
-        task_name=task_name,
-        query=MATH_PROMPT_TEMPLATE.format(prompt=line["question"]),
-        choices=[line["solution"]],
-        gold_index=0,
-    )
+def _maime_prompt_fn(lang: str):
+    template = MATH_PROMPT_TEMPLATES[lang]
+
+    def maime_prompt(line, task_name: str = None):
+        return Doc(
+            task_name=task_name,
+            query=template.format(prompt=line["question"]),
+            choices=[line["solution"]],
+            gold_index=0,
+        )
+
+    return maime_prompt
 
 
 # Danish tasks
 maime25_da = LightevalTaskConfig(
     name="maime25:da",
-    prompt_function=maime_prompt,
+    prompt_function=_maime_prompt_fn("da"),
     sample_fields=record_to_sample,
-    solver=[prompt_template(MATH_PROMPT_TEMPLATE), generate(cache=True)],
+    solver=[prompt_template(MATH_PROMPT_TEMPLATES["da"]), generate(cache=True)],
     scorer=math_scorer(),
     hf_repo="LumiOpen/mAIME2025",
     hf_subset="da_combined",
@@ -83,7 +97,7 @@ def maime_prompt(line, task_name: str = None):
 
 maime25_da_avg = LightevalTaskConfig(
     name="maime25_avg:da",
-    prompt_function=maime_prompt,
+    prompt_function=_maime_prompt_fn("da"),
     sample_fields=record_to_sample,
     hf_repo="LumiOpen/mAIME2025",
     hf_subset="da_combined",
@@ -98,7 +112,7 @@ def maime_prompt(line, task_name: str = None):
 
 maime25_da_gpassk = LightevalTaskConfig(
     name="maime25_gpassk:da",
-    prompt_function=maime_prompt,
+    prompt_function=_maime_prompt_fn("da"),
     sample_fields=record_to_sample,
     hf_repo="LumiOpen/mAIME2025",
     hf_subset="da_combined",
@@ -114,9 +128,9 @@ def maime_prompt(line, task_name: str = None):
 # Finnish tasks
 maime25_fi = LightevalTaskConfig(
     name="maime25:fi",
-    prompt_function=maime_prompt,
+    prompt_function=_maime_prompt_fn("fi"),
     sample_fields=record_to_sample,
-    solver=[prompt_template(MATH_PROMPT_TEMPLATE), generate(cache=True)],
+    solver=[prompt_template(MATH_PROMPT_TEMPLATES["fi"]), generate(cache=True)],
     scorer=math_scorer(),
     hf_repo="LumiOpen/mAIME2025",
     hf_subset="fi_combined",
@@ -134,7 +148,7 @@ def maime_prompt(line, task_name: str = None):
 
 maime25_fi_avg = LightevalTaskConfig(
     name="maime25_avg:fi",
-    prompt_function=maime_prompt,
+    prompt_function=_maime_prompt_fn("fi"),
     sample_fields=record_to_sample,
     hf_repo="LumiOpen/mAIME2025",
     hf_subset="fi_combined",
@@ -149,7 +163,7 @@ def maime_prompt(line, task_name: str = None):
 
 maime25_fi_gpassk = LightevalTaskConfig(
     name="maime25_gpassk:fi",
-    prompt_function=maime_prompt,
+    prompt_function=_maime_prompt_fn("fi"),
     sample_fields=record_to_sample,
     hf_repo="LumiOpen/mAIME2025",
     hf_subset="fi_combined",