evalscope 0.8.2__py3-none-any.whl → 0.10.0__py3-none-any.whl

evalscope/benchmarks/race/race_adapter.py

@@ -1,11 +1,12 @@
 # Copyright (c) Alibaba, Inc. and its affiliates.
 
-import json
 import os
 
-from evalscope.benchmarks.data_adapter import DataAdapter
-from evalscope.metrics.metrics import exact_match, weighted_mean
-from evalscope.utils import normalize_score
+from evalscope.benchmarks import Benchmark, DataAdapter
+from evalscope.constants import EvalType
+from evalscope.metrics import AverageAccuracy, exact_match
+from evalscope.models import MultiChoiceModelAdapter
+from evalscope.utils import ResponseParser
 from evalscope.utils.io_utils import jsonl_to_list
 from evalscope.utils.logger import get_logger
 
@@ -13,46 +14,28 @@ from evalscope.utils.logger import get_logger
 
 logger = get_logger()
 
-DATASET_ID = 'modelscope/race'
-
-SUBSET_LIST = ['high', 'middle']
-
-SUBJECT_MAPPING = {'high': 'High', 'middle': 'Middle'}
-
 
+@Benchmark.register(
+    name='race',
+    dataset_id='modelscope/race',
+    model_adapter=MultiChoiceModelAdapter,
+    subset_list=['high', 'middle'],
+    metric_list=[AverageAccuracy],
+    few_shot_num=3,
+    train_split='train',
+    eval_split='test',
+)
 class RACEAdapter(DataAdapter):
 
     choices = ['A', 'B', 'C', 'D']
 
-    def __init__(self,
-                 subset_list: list = None,
-                 metric_list: list = None,
-                 few_shot_num: int = None,
-                 train_split: str = 'train',
-                 eval_split: str = 'test',
-                 **kwargs):
-
-        if subset_list is None:
-            subset_list = SUBSET_LIST
-
-        if metric_list is None:
-            metric_list = [{'name': 'WeightedAverageAccuracy', 'object': weighted_mean}]
-
-        if few_shot_num is None:
-            logger.info(f'Set 3-shot examples by system for RACE.')
-            few_shot_num = 3
-
+    def __init__(self, **kwargs):
+        few_shot_num = kwargs.get('few_shot_num', 3)
         if few_shot_num > 3:
             logger.warning(f'few_shot_num <= 3 for RACE, but got {few_shot_num}. Use 3-shot by default.')
-            few_shot_num = 3
+            kwargs['few_shot_num'] = 3
 
-        super().__init__(
-            subset_list=subset_list,
-            metric_list=metric_list,
-            few_shot_num=few_shot_num,
-            train_split=train_split,
-            eval_split=eval_split,
-            **kwargs)
+        super().__init__(**kwargs)
 
     def load_from_disk(self, dataset_name_or_path, subset_list, work_dir, **kwargs) -> dict:
         data_dict = {}
@@ -99,13 +82,13 @@ class RACEAdapter(DataAdapter):
 
         full_prompt: str = context.strip() + self._generate_prompt(input_d=input_d, include_answer=False)
 
-        return {'data': [full_prompt], 'multi_choices': self.choices}
+        return {'data': [full_prompt], 'multi_choices': self.choices, 'system_prompt': self.prompt_template}
 
     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 = 'checkpoint') -> str:
+    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.
 
@@ -117,98 +100,18 @@ class RACEAdapter(DataAdapter):
         Returns:
             The parsed answer. Depending on the dataset. Usually a string for chat.
         """
-        if eval_type == 'checkpoint':
-            return result
-        elif eval_type == 'service':  # TODO: to be implemented
-            return result
-        elif eval_type == 'custom':  # TODO: to be implemented
+        if eval_type == EvalType.CHECKPOINT:
             return result
+        elif eval_type == EvalType.SERVICE:
+            return ResponseParser.parse_first_option_with_choices(result, self.choices)  # TODO: to be checked !
+        elif eval_type == EvalType.CUSTOM:
+            return ResponseParser.parse_first_option_with_choices(result, self.choices)  # TODO: to be checked !
         else:
             raise ValueError(f'Unknown eval_type: {eval_type}')
 
     def match(self, gold: str, pred: str) -> float:
         return exact_match(gold=gold, pred=pred)
 
-    def compute_metric(self, review_res_list: list) -> float:
-        """
-        Compute evaluation result by specific metric.
-
-        Args:
-            review_res_list: review score list, e.g. [0, 1, 1, 0, ...]
-
-        Returns:
-            The metric score.
-        """
-        items = [(score, 1.0) for score in review_res_list]
-        return weighted_mean(items)
-
-    def gen_report(self, subset_score_map: dict, report_name: str = None) -> dict:
-        """
-        Generate report for the evaluation.
-
-        Args:
-            subset_score_map: The subset-score mapping. e.g. {subset_name: (score, num), ...}
-            report_name: The user-defined report name.
-
-        Returns:
-        {
-            "name":"RACE",
-            "metric":"WeightedAverageAccuracy",
-            "score":0.3389,
-            "category":[
-                {
-                   "name":"High",
-                   "score":0.2528,
-                   "subset":[
-                       {
-                           "name":"high",
-                           "score":0.2528
-                       }
-                   ]
-                }
-            ],
-            "total_num":59
-        }
-        """
-        total_num: int = sum([num for _, num in subset_score_map.values()])
-        weighted_avg_acc: float = sum([score * num for score, num in subset_score_map.values()]) / total_num
-
-        # Get domain-subject mapping
-        subject_review_map = {}
-        for subset_name, (subset_score, num) in subset_score_map.items():
-            domain_name: str = SUBJECT_MAPPING.get(subset_name)
-            if domain_name in subject_review_map:
-                subject_review_map[domain_name].append((subset_name, subset_score, num))
-            else:
-                subject_review_map[domain_name] = [(subset_name, subset_score, num)]
-
-        # Get domain score
-        category_list = []
-        for domain_name, domain_res_list in subject_review_map.items():
-            domain_weighted_avg_acc = sum([score * num for _, score, num in domain_res_list]) / \
-                                     sum([num for _, _, num in domain_res_list])
-            domain_weighted_avg_acc = normalize_score(score=domain_weighted_avg_acc)
-            category_list.append({
-                'name':
-                domain_name,
-                'score':
-                normalize_score(score=domain_weighted_avg_acc),
-                'subset': [{
-                    'name': subset_name,
-                    'score': subset_score
-                } for subset_name, subset_score, _ in domain_res_list]
-            })
-
-        # Get final dict of report
-        res_map = dict(
-            name=report_name or 'race',
-            metric=self.metric_list[0]['name'],
-            score=weighted_avg_acc,
-            category=category_list,
-            total_num=total_num)
-
-        return res_map
-
     @classmethod
     def _generate_prompt(cls, input_d: dict, include_answer=True) -> str:
 

evalscope/benchmarks/trivia_qa/__init__.py

@@ -1,6 +1 @@
 # Copyright (c) Alibaba, Inc. and its affiliates.
-
-from evalscope.benchmarks.trivia_qa.trivia_qa_adapter import DATASET_ID, SUBSET_LIST
-from evalscope.benchmarks.trivia_qa.trivia_qa_adapter import TriviaQaAdapter
-from evalscope.benchmarks.trivia_qa.trivia_qa_adapter import TriviaQaAdapter as DataAdapterClass
-from evalscope.models.model_adapter import ChatGenerationModelAdapter as ModelAdapterClass  # noqa

evalscope/benchmarks/trivia_qa/trivia_qa_adapter.py

@@ -1,49 +1,35 @@
 # Copyright (c) Alibaba, Inc. and its affiliates.
 # Copyright (c) EleutherAI Inc, and its affiliates.
 import csv
-import numpy as np
 import os
-from typing import List
 
+from evalscope.benchmarks import Benchmark
 from evalscope.benchmarks.data_adapter import DataAdapter
-from evalscope.metrics.metrics import exact_match, weighted_mean
-from evalscope.utils.logger import get_logger
+from evalscope.constants import EvalType
+from evalscope.metrics import AverageAccuracy
+from evalscope.models import ChatGenerationModelAdapter
+from evalscope.utils import get_logger
 
 # flake8: noqa
 
 logger = get_logger()
 
-DATASET_ID = 'modelscope/trivia_qa'
-SUBSET_LIST = ['default']
-
 
+@Benchmark.register(
+    name='trivia_qa',
+    dataset_id='modelscope/trivia_qa',
+    model_adapter=ChatGenerationModelAdapter,
+    subset_list=['default'],
+    metric_list=[AverageAccuracy],
+    few_shot_num=5,
+    train_split='dev',
+    eval_split='test',
+)
 class TriviaQaAdapter(DataAdapter):
 
-    def __init__(self,
-                 subset_list: list = None,
-                 metric_list: list = None,
-                 few_shot_num: int = None,
-                 train_split: str = 'dev',
-                 eval_split: str = 'test',
-                 **kwargs):
-
-        if subset_list is None:
-            subset_list = SUBSET_LIST
-
-        if metric_list is None:
-            metric_list = [{'name': 'WeightedAverageAccuracy', 'object': weighted_mean}]
+    def __init__(self, **kwargs):
 
-        if few_shot_num is None:
-            logger.info(f'few_shot_num is not specified for TriviaQA, use default value: 5')
-            few_shot_num = 5
-
-        super().__init__(
-            subset_list=subset_list,
-            metric_list=metric_list,
-            few_shot_num=few_shot_num,
-            train_split=train_split,
-            eval_split=eval_split,
-            **kwargs)
+        super().__init__(**kwargs)
 
     def load_from_disk(self, dataset_name_or_path, subset_list, work_dir, **kwargs) -> dict:
         data_dict = {}
@@ -113,16 +99,16 @@ class TriviaQaAdapter(DataAdapter):
         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)
-        full_prompt = prompt + context
+        full_prompt = context
 
-        return {'data': [full_prompt]}
+        return {'data': [full_prompt], 'system_prompt': prompt or self.prompt_template}
 
     def get_gold_answer(self, input_d: dict) -> list:
         # Get the gold choice
         ans: list = input_d.get('ideal', [])
         return ans
 
-    def parse_pred_result(self, result: str, raw_input_d: dict = None, eval_type: str = 'checkpoint') -> str:
+    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.
 
@@ -134,73 +120,11 @@ class TriviaQaAdapter(DataAdapter):
         Returns:
             The predicted answer.
         """
-        if eval_type == 'checkpoint':
-            return result
-        elif eval_type == 'service':  # TODO: to be implemented
-            return result
-        elif eval_type == 'custom':  # TODO: to be implemented
-            return result
-        else:
-            raise ValueError(f'Unknown eval_type: {eval_type}')
+        return result
 
     def match(self, gold: list, pred: str) -> float:
-        return max([exact_match(gold=ref, pred=pred) for ref in gold])
-
-    def compute_metric(self, review_res_list: list) -> float:
-        """
-        Compute evaluation result by specific metric.
-
-        Args:
-            review_res_list: review score list, e.g. [0, 1, 1, 0, ...]
-
-        Returns:
-            The metric score.
-        """
-        items = [(score, 1.0) for score in review_res_list]
-        return weighted_mean(items)
-
-    def gen_report(self, subset_score_map: dict, report_name: str = None) -> dict:
-        """
-        Generate the report for the model output.
-
-        Args:
-            subset_score_map: {subset_name: (score, num), ...}
-            report_name: The user-defined report name.
-
-        Returns:
-        {
-            "name":"TriviaQA",
-            "metric":"WeightedAverageAccuracy",
-            "score":0.3389,
-            "category":[
-                {
-                    "name":"DEFAULT",
-                    "score":0.3389,
-                    "subset":[
-                        {
-                            "name":"default",
-                            "score":0.3389
-                        }
-                    ]
-                }
-            ],
-            "total_num":100
-        }
-        """
-        total_num: int = sum([num for _, num in subset_score_map.values()])
-        weighted_avg_acc: float = sum([score * num for score, num in subset_score_map.values()]) / total_num
-        cate_avg_list = [{'name': subset_name, 'score': score} for subset_name, (score, _) in subset_score_map.items()]
-
-        category_d = dict(name='DEFAULT', score=weighted_avg_acc, subset=cate_avg_list)
-
-        res_map = dict(
-            name=report_name or 'trivia_qa',
-            metric=self.metric_list[0]['name'],
-            score=weighted_avg_acc,
-            category=[category_d],
-            total_num=total_num)
-
-        return res_map
+        is_correct = any([cand in pred for cand in gold])
+        return 1 if is_correct else 0
 
     @classmethod
     def _generate_prompt(cls, input_d: dict, include_answer=True) -> str:

evalscope/benchmarks/truthful_qa/__init__.py

@@ -1,6 +1 @@
 # Copyright (c) Alibaba, Inc. and its affiliates.
-
-from evalscope.benchmarks.truthful_qa.truthful_qa_adapter import DATASET_ID, SUBSET_LIST
-from evalscope.benchmarks.truthful_qa.truthful_qa_adapter import TruthfulQaAdapter
-from evalscope.benchmarks.truthful_qa.truthful_qa_adapter import TruthfulQaAdapter as DataAdapterClass
-from evalscope.models.model_adapter import ContinuationLogitsModelAdapter as ModelAdapterClass  # noqa

evalscope/benchmarks/truthful_qa/truthful_qa_adapter.py

@@ -6,20 +6,30 @@ import numpy as np
 import os
 from typing import List
 
+from evalscope.benchmarks import Benchmark
 from evalscope.benchmarks.data_adapter import DataAdapter
-from evalscope.metrics.metrics import weighted_mean
-from evalscope.utils import normalize_score
-from evalscope.utils.logger import get_logger
+from evalscope.constants import EvalType
+from evalscope.metrics import AverageAccuracy
+from evalscope.models import ContinuationLogitsModelAdapter
+from evalscope.utils import get_logger, normalize_score
 
 # flake8: noqa
 
 logger = get_logger()
 
-DATASET_ID = 'modelscope/truthful_qa'
 # task_list = ['generation', 'multiple_choice']
-SUBSET_LIST = ['multiple_choice']
 
 
+@Benchmark.register(
+    name='truthful_qa',
+    dataset_id='modelscope/truthful_qa',
+    model_adapter=ContinuationLogitsModelAdapter,
+    subset_list=['multiple_choice'],
+    metric_list=[AverageAccuracy],
+    few_shot_num=0,
+    train_split=None,
+    eval_split='validation',
+)
 class TruthfulQaAdapter(DataAdapter):
     """
     Adapter for TruthfulQA benchmark.
@@ -40,35 +50,14 @@ class TruthfulQaAdapter(DataAdapter):
                  'Q: Where were the 1992 Olympics held?\n'
                  'A: The 1992 Olympics were held in Barcelona, Spain.')
 
-    def __init__(self,
-                 subset_list: list = None,
-                 metric_list: list = None,
-                 few_shot_num: int = None,
-                 train_split: str = None,
-                 eval_split: str = 'validation',
-                 **kwargs):
-
-        if subset_list is None:
-            subset_list = SUBSET_LIST
-
-        if metric_list is None:
-            metric_list = [{'name': 'WeightedAverageAccuracy', 'object': weighted_mean}]
-
-        if few_shot_num is None:
-            logger.info(f'Set 0-shot examples by system for TruthfulQA.')
-            few_shot_num = 0
+    def __init__(self, **kwargs):
 
+        few_shot_num = kwargs.get('few_shot_num', 0)
         if few_shot_num != 0:
             logger.warning(f'few_shot_num should be 0 for TruthfulQA, but got {few_shot_num}. Use 0-shot by default.')
-            few_shot_num = 0
+            kwargs['few_shot_num'] = 0
 
-        super().__init__(
-            subset_list=subset_list,
-            metric_list=metric_list,
-            few_shot_num=few_shot_num,
-            train_split=train_split,
-            eval_split=eval_split,
-            **kwargs)
+        super().__init__(**kwargs)
 
     def load_from_disk(self, dataset_name_or_path, subset_list, work_dir, **kwargs) -> dict:
         data_dict = {}
@@ -215,7 +204,7 @@ class TruthfulQaAdapter(DataAdapter):
         # TODO: generation sub-task to be added
         return {'mc1_labels': input_d['mc1_targets']['labels'], 'mc2_labels': input_d['mc2_targets']['labels']}
 
-    def parse_pred_result(self, result: list, raw_input_d: dict = None, eval_type: str = 'checkpoint') -> list:
+    def parse_pred_result(self, result: list, raw_input_d: dict = None, eval_type: str = EvalType.CHECKPOINT) -> list:
         """
         Parse the model output to get the answer.
 
@@ -227,11 +216,11 @@ class TruthfulQaAdapter(DataAdapter):
         Returns:
             The predicted answer.
         """
-        if eval_type == 'checkpoint':
+        if eval_type == EvalType.CHECKPOINT:
             return result
-        elif eval_type == 'service':  # TODO: to be supported !
+        elif eval_type == EvalType.SERVICE:  # TODO: to be supported !
             return result
-        elif eval_type == 'custom':  # TODO: to be supported !
+        elif eval_type == EvalType.CUSTOM:  # TODO: to be supported !
             return result
         else:
             raise ValueError(f'Invalid eval_type: {eval_type}')
@@ -270,7 +259,7 @@ class TruthfulQaAdapter(DataAdapter):
 
         return {'multiple_choice': {'mc1': mc1(mc1_lls), 'mc2': mc2(mc2_lls)}}  # or {'generation': xxx}
 
-    def compute_metric(self, review_res_list: List[dict]) -> float:
+    def compute_metric(self, review_res_list: List[dict]) -> List[dict]:
         """
         Compute evaluation result by specific metric for each subset.
 
@@ -295,56 +284,8 @@ class TruthfulQaAdapter(DataAdapter):
                 logger.error(f'** Unknown review_res: {review_res_d}')
 
         # To get mc2 score
-        items = [(score, 1.0) for score in mc2_list]
-        return weighted_mean(items)
-
-    def gen_report(self, subset_score_map: dict, report_name: str = None) -> dict:
-        """
-        Generate the report for the model output.
-
-        Args:
-            subset_score_map: {subset_name: (score, num), ...}
-            report_name: The user-defined report name.
-
-        Returns:
-        {
-            "name":"TruthfulQA",
-            "metric":"WeightedAverageAccuracy",
-            "score":0.3389,
-            "category":[
-                {
-                    "name":"DEFAULT",
-                    "score":0.2527,
-                    "subset":[
-                        {
-                            "name":"multiple_choice",
-                            "score":0.3157
-                        },
-                        # {
-                        #     "name":"generation",
-                        #     "score":0.2631
-                        # }
-                    ]
-                }
-            ],
-            "total_num":100
-        }
-        """
-        total_num: int = sum([num for _, num in subset_score_map.values()])
-        weighted_avg_acc: float = sum([score * num for score, num in subset_score_map.values()]) / total_num
-        weighted_avg_acc = normalize_score(score=weighted_avg_acc)
-        cate_avg_list = [{
-            'name': subset_name,
-            'score': normalize_score(score=score)
-        } for subset_name, (score, _) in subset_score_map.items()]
-
-        category_d = dict(name='DEFAULT', score=weighted_avg_acc, subset=cate_avg_list)
-
-        res_map = dict(
-            name=report_name or 'truthful_qa',
-            metric=self.metric_list[0]['name'],
-            score=weighted_avg_acc,
-            category=[category_d],
-            total_num=total_num)
-
-        return res_map
+        return [{
+            'metric_name': self.metric_list[0].name,
+            'score': self.metric_list[0].object(mc2_list),
+            'num': len(mc2_list)
+        }]

evalscope/cli/cli.py

@@ -2,6 +2,7 @@
 
 import argparse
 
+from evalscope.cli.start_app import StartAppCMD
 from evalscope.cli.start_eval import EvalCMD
 from evalscope.cli.start_perf import PerfBenchCMD
 
@@ -12,6 +13,7 @@ def run_cmd():
 
     PerfBenchCMD.define_args(subparsers)
     EvalCMD.define_args(subparsers)
+    StartAppCMD.define_args(subparsers)
 
     args = parser.parse_args()
 

evalscope/cli/start_app.py

@@ -0,0 +1,29 @@
+# Copyright (c) Alibaba, Inc. and its affiliates.
+import os
+from argparse import ArgumentParser
+
+from evalscope.cli.base import CLICommand
+from evalscope.report.app import create_app
+
+
+def subparser_func(args):
+    """ Function which will be called for a specific sub parser.
+    """
+    return StartAppCMD(args)
+
+
+class StartAppCMD(CLICommand):
+    name = 'app'
+
+    def __init__(self, args):
+        self.args = args
+
+    @staticmethod
+    def define_args(parsers: ArgumentParser):
+        """ define args for create pipeline template command.
+        """
+        parser = parsers.add_parser(StartAppCMD.name)
+        parser.set_defaults(func=subparser_func)
+
+    def execute(self):
+        create_app()

evalscope/collections/__init__.py

@@ -0,0 +1,3 @@
+from evalscope.collections.evaluator import EvaluatorCollection
+from evalscope.collections.sampler import StratifiedSampler, UniformSampler, WeightedSampler
+from evalscope.collections.schema import CollectionSchema, DatasetInfo