evalscope 0.10.1__py3-none-any.whl → 0.11.0__py3-none-any.whl

evalscope/benchmarks/general_mcq/general_mcq_adapter.py

@@ -0,0 +1,129 @@
+# Copyright (c) Alibaba, Inc. and its affiliates.
+import csv
+import os
+
+from evalscope.benchmarks import Benchmark, DataAdapter
+from evalscope.constants import EvalType
+from evalscope.metrics.metrics import exact_match
+from evalscope.models import MultiChoiceModelAdapter
+from evalscope.utils import ResponseParser
+from evalscope.utils.logger import get_logger
+
+# flake8: noqa
+
+logger = get_logger()
+
+
+@Benchmark.register(
+    name='general_mcq',
+    dataset_id='general_mcq',
+    model_adapter=MultiChoiceModelAdapter,
+    subset_list=['default'],
+    metric_list=['AverageAccuracy'],
+    few_shot_num=0,
+    train_split='dev',
+    eval_split='val',
+    prompt_template='请回答问题，并选出其中的正确答案\n{query}',
+)
+class GeneralMCQAdapter(DataAdapter):
+
+    choices = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J']
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+
+    def load_from_disk(self, dataset_name_or_path, subset_list, work_dir, **kwargs) -> dict:
+        data_dict = {}
+        for subset_name in subset_list:
+            for split_name in [self.train_split, self.eval_split]:
+                if os.path.exists(dataset_name_or_path):
+                    file_path = os.path.join(dataset_name_or_path, f'{subset_name}_{split_name}.csv')
+                else:
+                    file_path = os.path.join(work_dir, dataset_name_or_path, f'{subset_name}_{split_name}.csv')
+                if os.path.exists(file_path):
+                    with open(file_path, encoding='utf-8') as f:
+                        rows = []
+                        reader = csv.reader(f)
+                        header = next(reader)
+                        for row in reader:
+                            item = dict(zip(header, row))
+                            rows.append(item)
+
+                        if subset_name in data_dict:
+                            data_dict[subset_name].update({split_name: rows})
+                        else:
+                            data_dict[subset_name] = {split_name: rows}
+
+        return data_dict
+
+    def gen_prompt(self, input_d: dict, subset_name: str, few_shot_list: list, **kwargs) -> dict:
+        """
+        Generate model prompt from raw input, unify the prompt format for C-Eval benchmark.
+
+        Args:
+            input_d (dict): The raw input. A single data format of the C-Eval:
+
+            {'id': 0,
+            'question': '下列关于税法基本原则的表述中，不正确的是____。',
+            'A': '税收法定原则包括税收要件法定原则和税务合法性原则',
+            'B': '税收公平原则源于法律上的平等性原则',
+            'C': '税收效率原则包含经济效率和行政效率两个方面',
+            'D': '税务机关按法定程序依法征税，可以自由做出减征、停征或免征税款的决定',
+            'answer': 'D'}
+
+        Returns:
+            {'data': ['prompt ...']}
+        """
+
+        few_shot_prompts = [self._format_example(input_d=sample, include_answer=True) for sample in few_shot_list]
+
+        if len(few_shot_prompts) > 0:
+            context: str = '\n'.join(few_shot_prompts) + '\n'
+        else:
+            context = ''
+        context = context.strip() + self._format_example(input_d=input_d, include_answer=False)
+
+        full_prompt = self.prompt_template.format(query=context)
+
+        return {'data': [full_prompt], 'multi_choices': self.choices, 'system_prompt': self.system_prompt}
+
+    def get_gold_answer(self, input_d: dict) -> str:
+        # Get the gold choice
+        return input_d.get('answer', '')
+
+    def parse_pred_result(self, result: str, raw_input_d: dict = None, eval_type: str = EvalType.CHECKPOINT) -> str:
+        """
+        Parse the model output to get the answer. Could be the best choice index.
+
+        Args:
+            result: Predicted answer from the model. Usually a string for chat.
+            raw_input_d (dict): The raw input. Depending on the dataset.
+            eval_type: `checkpoint` or `service` or `custom`. Default is `checkpoint`.
+
+        Returns:
+            The parsed answer. Depending on the dataset. Usually a string for chat.
+        """
+        if eval_type == EvalType.CHECKPOINT:
+            return result
+        elif eval_type == EvalType.SERVICE:
+            return ResponseParser.parse_first_option_with_choices(result, self.choices)
+        elif eval_type == EvalType.CUSTOM:
+            return ResponseParser.parse_first_option_with_choices(result, self.choices)
+        else:
+            raise ValueError(f'Invalid eval_type: {eval_type}')
+
+    def match(self, gold: str, pred: str) -> float:
+        return exact_match(gold=gold, pred=pred)
+
+    @classmethod
+    def _format_example(cls, input_d: dict, include_answer=True):
+        example = '问题：' + input_d['question']
+        for choice in cls.choices:
+            if choice in input_d:
+                example += f'\n{choice}. {input_d[f"{choice}"]}'
+
+        if include_answer:
+            example += '\n答案: ' + input_d['answer'] + '\n\n'
+        else:
+            example += '\n答案: '
+        return example

evalscope/benchmarks/general_qa/general_qa_adapter.py

@@ -5,7 +5,7 @@ from collections import defaultdict
 from typing import List
 
 from evalscope.benchmarks import Benchmark, DataAdapter
-from evalscope.metrics import AverageBLEU, bleu_ngram_one_sample, compute_rouge_score_one_sample_zh, mean
+from evalscope.metrics import bleu_ngram_one_sample, compute_rouge_score_one_sample_zh, mean
 from evalscope.models import ChatGenerationModelAdapter
 from evalscope.utils.io_utils import jsonl_to_list
 from evalscope.utils.logger import get_logger
@@ -18,7 +18,7 @@ logger = get_logger()
     dataset_id='general_qa',
     model_adapter=ChatGenerationModelAdapter,
     subset_list=['default'],
-    metric_list=[AverageBLEU],
+    metric_list=['AverageBLEU'],
     few_shot_num=0,
     train_split=None,
     eval_split='test',
@@ -30,16 +30,16 @@ class GeneralQAAdapter(DataAdapter):
 
         super().__init__(**kwargs)
 
-    def load(self, dataset_name_or_path: str, subset_list: list = None, **kwargs) -> dict:
+    def load(self, **kwargs) -> dict:
 
-        data_file_list = glob.glob(os.path.join(dataset_name_or_path, '*.jsonl'))
+        data_file_list = glob.glob(os.path.join(self.dataset_id, '*.jsonl'))
         data_list = []
 
         try:
             for file_path in data_file_list:
                 data_list.extend(jsonl_to_list(file_path))
         except Exception as e:
-            raise ValueError(f'Failed to load data from {dataset_name_or_path}, got error: {e}')
+            raise ValueError(f'Failed to load data from {self.dataset_id}, got error: {e}')
 
         data_dict = {'default': {'test': data_list}}
 
@@ -66,7 +66,7 @@ class GeneralQAAdapter(DataAdapter):
 
         # if len(history) > 0:
         #     prompt = '\n'.join(history) + '\n' + prompt
-        return {'data': [prompt], 'system_prompt': self.prompt_template}
+        return {'data': [prompt], 'system_prompt': self.system_prompt}
 
     def get_gold_answer(self, input_d: dict) -> str:
         """

evalscope/benchmarks/gpqa/gpqa_adapter.py

@@ -3,10 +3,9 @@ import random
 import re
 
 from evalscope.benchmarks import Benchmark, DataAdapter
-from evalscope.constants import AnswerKeys, EvalType
-from evalscope.metrics import Pass1, exact_match
+from evalscope.constants import EvalType
+from evalscope.metrics import exact_match
 from evalscope.models import ChatGenerationModelAdapter
-from evalscope.utils.utils import ResponseParser
 
 
 @Benchmark.register(
@@ -14,11 +13,11 @@ from evalscope.utils.utils import ResponseParser
     dataset_id='modelscope/gpqa',
     model_adapter=ChatGenerationModelAdapter,
     subset_list=['gpqa_extended', 'gpqa_main', 'gpqa_diamond'],
-    metric_list=[Pass1],
+    metric_list=['AveragePass@1'],
     few_shot_num=5,
     train_split='train',
     eval_split='train',  # only have train split
-    prompt_template='',
+    prompt_template='{query}\nPlease reason step by step, and put your final answer within \\boxed{{}}.',
 )
 class GPQAAdapter(DataAdapter):
 
@@ -48,9 +47,10 @@ class GPQAAdapter(DataAdapter):
         """  # noqa: E501
         processed_input_d = self.__process_input(input_d)
         input_d['answer'] = processed_input_d['answer']  # add answer to input_d for answer extraction
-        prompt = self.prompt_prefix + f"{input_d['Question']}\n{self.__form_options(processed_input_d['choices'])}Let's think step by step: "  # noqa: E501
+        query = self.prompt_prefix + f"{input_d['Question']}\n{self.__form_options(processed_input_d['choices'])}"  # noqa: E501
 
-        return {'data': [prompt], 'multi_choices': self.choices, 'system_prompt': self.prompt_template}
+        prompt = self.prompt_template.format(query=query)
+        return {'data': [prompt], 'multi_choices': self.choices, 'system_prompt': self.system_prompt}
 
     def __process_input(self, input_d: dict) -> dict:
 
@@ -94,10 +94,28 @@ class GPQAAdapter(DataAdapter):
         """
         Parse the predicted result and extract proper answer.
         """
-        return ResponseParser.parse_first_option_with_choices(result, self.choices)
+        return GPQAAdapter.get_multiple_choice_answer(result)
 
     def match(self, gold: str, pred: str) -> float:
         """
         Match the gold answer and the predicted answer.
         """
         return exact_match(gold=gold, pred=pred)
+
+    @staticmethod
+    def get_multiple_choice_answer(pred: str):
+        tmp = re.findall(r'\b(A|B|C|D)\b', pred.upper())
+        if tmp:
+            pred = tmp
+        else:
+            pred = [pred.strip().strip('.')]
+
+        if len(pred) == 0:
+            pred = ''
+        else:
+            pred = pred[-1]
+
+        # Remove the period at the end, again!
+        pred = pred.rstrip('.').rstrip('/')
+
+        return pred

evalscope/benchmarks/gsm8k/gsm8k_adapter.py

@@ -6,7 +6,6 @@ import os
 import re
 
 from evalscope.benchmarks import Benchmark, DataAdapter
-from evalscope.metrics import AverageAccuracy
 from evalscope.models import ChatGenerationModelAdapter
 from evalscope.utils.io_utils import jsonl_to_list
 from evalscope.utils.logger import get_logger
@@ -19,11 +18,11 @@ logger = get_logger()
     dataset_id='modelscope/gsm8k',
     model_adapter=ChatGenerationModelAdapter,
     subset_list=['main'],
-    metric_list=[AverageAccuracy],
+    metric_list=['AverageAccuracy'],
     few_shot_num=4,
     train_split='train',
     eval_split='test',
-    prompt_template='',
+    prompt_template="Question: {query}\nLet's think step by step\nAnswer:",
 )
 class GSM8KAdapter(DataAdapter):
 
@@ -73,10 +72,11 @@ class GSM8KAdapter(DataAdapter):
             }
         """
         use_fewshot = self.few_shot_num > 0
+        context = self._generate_prompt(use_fewshot=use_fewshot)
 
-        full_prompt = self._generate_prompt(input_d, few_shot_list=few_shot_list, use_fewshot=use_fewshot)
+        full_prompt = context + self.prompt_template.format(query=input_d['question'])
 
-        return {'data': [full_prompt], 'system_prompt': self.prompt_template}
+        return {'data': [full_prompt], 'system_prompt': self.system_prompt}
 
     def get_gold_answer(self, input_d: dict) -> str:
         # Extract the gold answer from the input dict.
@@ -123,7 +123,7 @@ class GSM8KAdapter(DataAdapter):
         return number_equal(gold_ans=gold, pred_ans=pred)
 
     @classmethod
-    def _generate_prompt(cls, input_d: dict, few_shot_list: list, use_fewshot: bool = True) -> str:
+    def _generate_prompt(cls, use_fewshot: bool = True) -> str:
         if use_fewshot:
             # Use 4-shot examples by system
             context = (
@@ -135,14 +135,9 @@ class GSM8KAdapter(DataAdapter):
                 "When Bella buys 2/5 times more marbles, she'll have increased the number of marbles by 2/5*60 = 24\nThe total number of marbles she'll have is 60+24 = 84\nIf Bella currently has 60 marbles, and she has two times as many marbles as frisbees, she has 60/2 = 30 frisbees.\nIf Bella buys 2/5 times more frisbees, she'll have 2/5*30 = 12 more frisbees.\nThe total number of frisbees she'll have will increase to 30+12 = 42\nBella also has 20 more frisbees than deck cards, meaning she has 30-20 = 10 deck cards\nIf she buys 2/5 times more deck cards, she'll have 2/5*10 = 4 more deck cards.\nThe total number of deck cards she'll have is 10+4 = 14\nTogether, Bella will have a total of 14+42+84 = 140 items\nThe answer is 140\n\n"
                 "Question: A group of 4 fruit baskets contains 9 apples, 15 oranges, and 14 bananas in the first three baskets and 2 less of each fruit in the fourth basket. How many fruits are there?\nLet's think step by step\n"
                 'For the first three baskets, the number of apples and oranges in one basket is 9+15=24\nIn total, together with bananas, the number of fruits in one basket is 24+14=38 for the first three baskets.\nSince there are three baskets each having 38 fruits, there are 3*38=114 fruits in the first three baskets.\nThe number of apples in the fourth basket is 9-2=7\nThere are also 15-2=13 oranges in the fourth basket\nThe combined number of oranges and apples in the fourth basket is 13+7=20\nThe fourth basket also contains 14-2=12 bananas.\nIn total, the fourth basket has 20+12=32 fruits.\nThe four baskets together have 32+114=146 fruits.\nThe answer is 146\n\n'
-                f"Question: {input_d['question']}\nLet's think step by step\nAnswer:")
-            # context = input_d['question']
-            # fewshot_prompts = ['Question: ' + item_d['question'] + '\nAnswer: ' + item_d['answer'] for item_d in few_shot_list]
-            # fewshot_prompts = fewshot_prompts + ['Question: ' + context + '\nAnswer:']
-            # context = '\n\n'.join(fewshot_prompts)
+            )
         else:
-            context = input_d['question']
-            context = 'Question: ' + context + '\nAnswer:'
+            context = ''
         return context
 
     @staticmethod

evalscope/benchmarks/hellaswag/hellaswag_adapter.py

@@ -5,7 +5,7 @@ import re
 
 from evalscope.benchmarks import Benchmark, DataAdapter
 from evalscope.constants import EvalType
-from evalscope.metrics import AverageAccuracy, exact_match
+from evalscope.metrics import exact_match
 from evalscope.models import ContinuationLogitsModelAdapter
 from evalscope.utils.io_utils import jsonl_to_list
 from evalscope.utils.logger import get_logger
@@ -21,7 +21,7 @@ logger = get_logger()
     dataset_id='modelscope/hellaswag',
     model_adapter=ContinuationLogitsModelAdapter,
     subset_list=['default'],
-    metric_list=[AverageAccuracy],
+    metric_list=['AverageAccuracy'],
     few_shot_num=0,
     train_split='train',
     eval_split='validation',
@@ -89,11 +89,7 @@ class HellaSwagAdapter(DataAdapter):
 
         ctx_continuation_pair_list = [(context.strip(), ' ' + cont.strip()) for cont in endings]
 
-        return {
-            'data': ctx_continuation_pair_list,
-            'multi_choices': self.choices,
-            'system_prompt': self.prompt_template
-        }
+        return {'data': ctx_continuation_pair_list, 'multi_choices': self.choices, 'system_prompt': self.system_prompt}
 
     def get_gold_answer(self, input_d: dict) -> str:
         # Get the gold choice

evalscope/benchmarks/humaneval/humaneval_adapter.py

@@ -2,7 +2,6 @@
 import re
 
 from evalscope.benchmarks import Benchmark, DataAdapter
-from evalscope.metrics import Pass1
 from evalscope.models import ChatGenerationModelAdapter
 from evalscope.utils.logger import get_logger
 
@@ -17,11 +16,11 @@ logger = get_logger()
     dataset_id='modelscope/humaneval',
     model_adapter=ChatGenerationModelAdapter,
     subset_list=['openai_humaneval'],
-    metric_list=[Pass1],
+    metric_list=['Pass@1'],
     few_shot_num=0,
     train_split=None,
     eval_split='test',
-    prompt_template='',
+    prompt_template='Complete the following python code:\n{query}',
 )
 class HumanevalAdapter(DataAdapter):
     """
@@ -64,10 +63,10 @@ class HumanevalAdapter(DataAdapter):
             input_d (dict): The raw input. A single data format of the Humaneval:
             {'task_id': '', 'prompt': '', 'entry_point': '', 'canonical_solution': '', 'test': ''}
         """
-        full_prompt = input_d['prompt']
-        full_prompt = f'Complete the following python code:\n{full_prompt}' if self.prompt_template else full_prompt
+        query = input_d['prompt']
+        full_prompt = self.prompt_template.format(query=query)
 
-        return {'data': [full_prompt], 'system_prompt': self.prompt_template}
+        return {'data': [full_prompt], 'system_prompt': self.system_prompt}
 
     @classmethod
     def _postprocess(cls, text: str) -> str:

evalscope/benchmarks/ifeval/ifeval_adapter.py

@@ -2,9 +2,9 @@ from collections import defaultdict
 from typing import Any, Dict, List
 
 from evalscope.benchmarks import Benchmark, DataAdapter
-from evalscope.benchmarks.ifeval.utils import agg_inst_level_acc, process_results
+from evalscope.benchmarks.ifeval.utils import process_results
 from evalscope.constants import EvalType
-from evalscope.metrics import Metric, mean
+from evalscope.metrics import Metric, mean, metric_registry
 from evalscope.models import ChatGenerationModelAdapter
 
 
@@ -14,10 +14,10 @@ from evalscope.models import ChatGenerationModelAdapter
     model_adapter=ChatGenerationModelAdapter,
     subset_list=['default'],
     metric_list=[
-        Metric(name='prompt_level_strict_acc', object=mean),
-        Metric(name='inst_level_strict_acc', object=agg_inst_level_acc),
-        Metric(name='prompt_level_loose_acc', object=mean),
-        Metric(name='inst_level_loose_acc', object=agg_inst_level_acc),
+        'prompt_level_strict_acc',
+        'inst_level_strict_acc',
+        'prompt_level_loose_acc',
+        'inst_level_loose_acc',
     ],
     few_shot_num=0,
     train_split=None,
@@ -29,8 +29,14 @@ class IFEvalAdapter(DataAdapter):
     def __init__(self, **kwargs):
         super().__init__(**kwargs)
 
+        # register metrics
+        metric_registry.register(Metric(name='prompt_level_strict_acc', object=mean))
+        metric_registry.register(Metric(name='inst_level_strict_acc', object=mean))
+        metric_registry.register(Metric(name='prompt_level_loose_acc', object=mean))
+        metric_registry.register(Metric(name='inst_level_loose_acc', object=mean))
+
     def gen_prompt(self, input_d: dict, subset_name: str, few_shot_list: list, **kwargs) -> Any:
-        return {'data': [input_d['prompt']], 'system_prompt': self.prompt_template}
+        return {'data': [input_d['prompt']], 'system_prompt': self.system_prompt}
 
     def get_gold_answer(self, input_d: dict) -> str:
         return input_d
@@ -48,9 +54,4 @@ class IFEvalAdapter(DataAdapter):
             for k, v in res.items():
                 res_dict[k].append(v)
 
-        metrics = []
-        for metric in self.metric_list:
-            metric_name = metric.name
-            pred_value = res_dict[metric_name]
-            metrics.append({'metric_name': metric_name, 'score': metric.object(pred_value), 'num': len(pred_value)})
-        return metrics
+        return super().compute_metric(res_dict)

evalscope/benchmarks/iquiz/iquiz_adapter.py

@@ -1,6 +1,6 @@
 from evalscope.benchmarks import Benchmark, DataAdapter
-from evalscope.constants import AnswerKeys, EvalType
-from evalscope.metrics import AverageAccuracy, exact_match
+from evalscope.constants import EvalType
+from evalscope.metrics import exact_match
 from evalscope.models import ChatGenerationModelAdapter
 from evalscope.utils.utils import ResponseParser
 
@@ -10,11 +10,11 @@ from evalscope.utils.utils import ResponseParser
     dataset_id='AI-ModelScope/IQuiz',
     model_adapter=ChatGenerationModelAdapter,
     subset_list=['IQ', 'EQ'],
-    metric_list=[AverageAccuracy],
+    metric_list=['AverageAccuracy'],
     few_shot_num=0,
     train_split=None,
     eval_split='test',
-    prompt_template='你是一个高智商和高情商的专家，你被要求回答一个选择题，并选出一个正确的选项，解释原因，最终输出格式为：`答案是(选项)`。',  # noqa: E501
+    system_prompt='你是一个高智商和高情商的专家，你被要求回答一个选择题，并选出一个正确的选项，解释原因，最终输出格式为：`答案是(选项)`。',  # noqa: E501
 )
 class IQuizAdapter(DataAdapter):
 
@@ -36,7 +36,7 @@ class IQuizAdapter(DataAdapter):
         """
         prompt = f"问题: {input_d['question']}\n"
         prompt += self.__form_options(input_d['choices'])
-        return {'data': [prompt], 'multi_choices': self.choices, 'system_prompt': self.prompt_template}
+        return {'data': [prompt], 'multi_choices': self.choices, 'system_prompt': self.system_prompt}
 
     def __form_options(self, options: list):
         option_str = '选项:\n'

evalscope/benchmarks/math_500/math_500_adapter.py

@@ -0,0 +1,49 @@
+from evalscope.benchmarks import Benchmark, DataAdapter
+from evalscope.metrics.math_parser import extract_answer, math_equal, strip_answer_string
+from evalscope.models import ChatGenerationModelAdapter
+from evalscope.utils.logger import get_logger
+
+# flake8: noqa
+
+logger = get_logger()
+
+
+@Benchmark.register(
+    name='math_500',
+    dataset_id='AI-ModelScope/MATH-500',
+    model_adapter=ChatGenerationModelAdapter,
+    subset_list=['default'],
+    metric_list=['AveragePass@1'],
+    few_shot_num=0,
+    train_split=None,
+    eval_split='test',
+    prompt_template='{query}\nPlease reason step by step, and put your final answer within \\boxed{{}}.',
+)
+class Math500Adapter(DataAdapter):
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+    def gen_prompt(self, input_d: dict, few_shot_list: list, **kwargs) -> dict:
+        """
+        Generate the prompt for the model input.
+        """
+        problem = input_d['problem']
+        full_prompt = self.prompt_template.format(query=problem)
+
+        return {'data': [full_prompt], 'system_prompt': self.system_prompt}
+
+    def get_gold_answer(self, input_d: dict) -> str:
+        # Extract the gold answer from the input dict.
+        return strip_answer_string(input_d['answer'])
+
+    def parse_pred_result(self, result: str, raw_input_d: dict = None, eval_type: str = 'checkpoint') -> str:
+        """
+        Parse the model output to get the answer. Could be the best choice index.
+        """
+        # Note: Use same extraction method for both of checkpoint/service/custom
+        result = strip_answer_string(extract_answer(result))
+        return result
+
+    def match(self, gold: str, pred: str) -> float:
+        return math_equal(pred, gold)

evalscope/benchmarks/mmlu/mmlu_adapter.py

@@ -4,17 +4,15 @@ import os
 
 from evalscope.benchmarks import Benchmark, DataAdapter
 from evalscope.constants import EvalType
-from evalscope.metrics import AverageAccuracy, exact_match
+from evalscope.metrics import exact_match
 from evalscope.models import MultiChoiceModelAdapter
-from evalscope.utils import ResponseParser, normalize_score
+from evalscope.utils import ResponseParser
 from evalscope.utils.logger import get_logger
 
 # flake8: noqa
 
 logger = get_logger()
 
-DATASET_ID = 'modelscope/mmlu'
-
 SUBSET_LIST = [
     'high_school_european_history',
     'business_ethics',
@@ -141,11 +139,11 @@ SUBJECT_MAPPING = {
     dataset_id='modelscope/mmlu',
     model_adapter=MultiChoiceModelAdapter,
     subset_list=SUBSET_LIST,
-    metric_list=[AverageAccuracy],
+    metric_list=['AverageAccuracy'],
     few_shot_num=5,
     train_split='train',
     eval_split='test',
-    prompt_template='',
+    prompt_template='The following are multiple choice questions (with answers) about {subset_name}. \n{query}',
 )
 class MMLUAdapter(DataAdapter):
 
@@ -221,17 +219,15 @@ class MMLUAdapter(DataAdapter):
             {'data': [full_prompt], 'multi_choices': self.choices}
 
         """
-        prompt = 'The following are multiple choice questions (with answers) about {}.\n\n'.format(
-            self._format_subject(subset_name))
         few_shot_prompts = [self._generate_prompt(input_d=sample, include_answer=True) for sample in few_shot_list]
 
         context: str = '\n'.join(few_shot_prompts) + '\n'
         context += self._generate_prompt(input_d=input_d, include_answer=False)
-        context = prompt + context
+        query = context.strip() + self._generate_prompt(input_d=input_d, include_answer=False)
 
-        full_prompt: str = context.strip() + self._generate_prompt(input_d=input_d, include_answer=False)
+        full_prompt = self.prompt_template.format(subset_name=self._format_subject(subset_name), query=query)
 
-        return {'data': [full_prompt], 'multi_choices': self.choices, 'system_prompt': self.prompt_template}
+        return {'data': [full_prompt], 'multi_choices': self.choices, 'system_prompt': self.system_prompt}
 
     def get_gold_answer(self, input_d: dict) -> str:
         # Get the gold choice

evalscope/benchmarks/mmlu_pro/mmlu_pro_adapter.py

@@ -3,22 +3,27 @@ from typing import Any, Dict
 
 from evalscope.benchmarks import Benchmark, DataAdapter
 from evalscope.constants import AnswerKeys, EvalType
-from evalscope.metrics import AverageAccuracy, exact_match
+from evalscope.metrics import exact_match
 from evalscope.models import ChatGenerationModelAdapter
 from evalscope.utils.utils import ResponseParser
 
+SUBSET_LIST = [
+    'computer science', 'math', 'chemistry', 'engineering', 'law', 'biology', 'health', 'physics', 'business',
+    'philosophy', 'economics', 'other', 'psychology', 'history'
+]
+
 
 @Benchmark.register(
     name='mmlu_pro',
     dataset_id='modelscope/mmlu-pro',
     model_adapter=ChatGenerationModelAdapter,
-    subset_list=['default'],
-    metric_list=[AverageAccuracy],
+    subset_list=SUBSET_LIST,
+    metric_list=['AverageAccuracy'],
     few_shot_num=5,
     train_split='validation',
     eval_split='test',
     prompt_template=
-    'You are an knowledge expert, you are supposed to answer the multi-choice question to derive your final answer as `The answer is ...`.',  # noqa: E501
+    'The following are multiple choice questions (with answers) about {subset_name}. Think step by step and then finish your answer with \"the answer is (X)\" where X is the correct letter choice.\n{query}',  # noqa: E501
 )
 class MMLUProAdapter(DataAdapter):
 
@@ -26,10 +31,11 @@ class MMLUProAdapter(DataAdapter):
         super().__init__(**kwargs)
 
         self.choices = ['A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J']
-        self.categories = [
-            'computer science', 'math', 'chemistry', 'engineering', 'law', 'biology', 'health', 'physics', 'business',
-            'philosophy', 'economics', 'other', 'psychology', 'history'
-        ]
+
+    def load(self, **kwargs):
+        # default load all data
+        kwargs['subset_list'] = ['default']
+        return super().load(**kwargs)
 
     def gen_prompts(self, data_dict: dict, **kwargs) -> Dict[str, list]:
         """
@@ -37,26 +43,32 @@ class MMLUProAdapter(DataAdapter):
         Return a dict with category as key and list of prompts as value.
         """
 
-        data_dict = data_dict[self.subset_list[0]]  # Only one subset for MMLU-Pro
+        data_dict = data_dict['default']  # Only one subset for MMLU-Pro
         fewshot_prompts = self.get_fewshot_examples(data_dict)
 
         #  Use the category as key to group the prompts
         res_dict = defaultdict(list)
         # generate prompts for each test sample
         for entry in data_dict[self.eval_split]:
-            prefix = fewshot_prompts[entry['category']]
+            subset_name = entry['category']
+            if subset_name not in self.subset_list:
+                continue
+            prefix = fewshot_prompts[subset_name]
             query = prefix + 'Q: ' + entry['question'] + '\n' + \
                 self.__form_options(entry['options']) + '\n'
 
-            prompt_d = {'data': [query], 'system_prompt': self.prompt_template, AnswerKeys.RAW_INPUT: entry}
+            full_prompt = self.prompt_template.format(subset_name=subset_name, query=query)
+            prompt_d = {'data': [full_prompt], 'system_prompt': self.system_prompt, AnswerKeys.RAW_INPUT: entry}
 
-            res_dict[entry['category']].append(prompt_d)
+            res_dict[subset_name].append(prompt_d)
         return res_dict
 
     def get_fewshot_examples(self, data_dict: dict):
-        # load 5-shot prompts for each category
-        prompts = {c: '' for c in self.categories}
-        for d in data_dict[self.train_split]:
+        # load few-shot prompts for each category
+        prompts = {c: '' for c in self.subset_list}
+        for index, d in enumerate(data_dict[self.train_split]):
+            if index >= self.few_shot_num:
+                break
             prompts[d['category']] += 'Q:' + ' ' + d['question'] + '\n' + \
                 self.__form_options(d['options']) + '\n' + \
                 d['cot_content'] + '\n\n'