[Sync] Merge branch 'dev' into zfz/update-keyset-demo (#876)

.pre-commit-config-zh-cn.yaml

@@ -7,6 +7,7 @@ exclude: |
       opencompass/datasets/lawbench/utils|
       opencompass/datasets/lawbench/evaluation_functions/|
       opencompass/datasets/medbench/|
+      opencompass/datasets/teval/|
       opencompass/datasets/NPHardEval/|
       docs/zh_cn/advanced_guides/compassbench_intro.md
     )

.pre-commit-config.yaml

@@ -7,6 +7,7 @@ exclude: |
       opencompass/datasets/lawbench/utils|
       opencompass/datasets/lawbench/evaluation_functions/|
       opencompass/datasets/medbench/|
+      opencompass/datasets/teval/|
       opencompass/datasets/NPHardEval/|
       docs/zh_cn/advanced_guides/compassbench_intro.md
     )

configs/datasets/MathBench/mathbench_2024_gen_de9ff9.py

+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import GenInferencer
+from opencompass.openicl.icl_evaluator import CircularEvaluator, AccEvaluator
+from opencompass.datasets import MathBenchDataset, mathbench_postprocess
+from opencompass.utils.text_postprocessors import first_option_postprocess
+
+single_choice_prompts = {
+    "single_choice_cn_with_reasoning": "以下是一道关于数学的单项选择题，请你一步一步推理，并在最后用“所以答案为选项X”给出答案，其中“X”为选项A，B，C，D中你认为正确的选项。下面是你要回答的问题\n{question}\n让我们一步一步思考：\n",
+    "single_choice_cn": "以下是一道关于数学的单项选择题，请你直接回答正确答案的选项序号。\n下面是你要回答的题目：\n{question}\n答案选项：",
+    "single_choice_en_with_reasoning": "Here is a multiple-choice question about mathematics. Please reason through it step by step, and at the end, provide your answer option with 'Therefore, the correct answer is option X', Where 'X' is the correct option you think from A，B，C，D. Here is the question you need to answer:\n{question}\nLet's think step by step:",
+    "single_choice_en": "Here is a multiple-choice question about mathematics. Please provide the correct answer option directly.\nHere is the question you need to answer:\n{question}\nAnswer option:",
+}
+
+cloze_prompts = {
+    "cloze_cn": [
+                dict(role='HUMAN', prompt='Q: 林中有15棵树。林务工人员今天将在林中种植树木。完成后，将有21棵树。林务工人员今天种植了多少棵树？'),
+                dict(role='BOT', prompt='A: 我们从15棵树开始。后来有21棵树。差值必定是他们种植的树木数量。所以，他们必须种植了21 - 15 = 6棵树。答案是 6\n'),
+                dict(role='HUMAN', prompt='Q: 如果停车场有3辆车，又有2辆车进来，停车场里有多少辆车？'),
+                dict(role='BOT', prompt='A: 停车场已经有3辆车。又进来了2辆车。现在有3 + 2 = 5辆车。答案是 5\n'),
+                dict(role='HUMAN', prompt='Q: 黎恩有32块巧克力，她的妹妹有42块。如果他们吃了35块，他们总共剩下多少块？'),
+                dict(role='BOT', prompt='A: 黎恩有32块巧克力，Leah的妹妹有42块。这意味着原本有32 + 42 = 74块巧克力。被吃掉了35块。所以他们总共还剩下74 - 35 = 39块巧克力。答案是 39\n'),
+                dict(role='HUMAN', prompt='Q: 杰森有20个棒棒糖。他给丹妮一些棒棒糖。现在Jason只剩下12个棒棒糖。杰森给丹妮多少个棒棒糖？'),
+                dict(role='BOT', prompt='A: 杰森有20个棒棒糖。因为他现在只剩下12个，所以他必须把剩下的都给了丹妮。他给丹妮的棒棒糖数量必定是20 - 12 = 8个。答案是 8\n'),
+                dict(role='HUMAN', prompt='Q: 莎莎有五个玩具。在圣诞节，他从他的爸爸和妈妈那里各得到了两个玩具。现在他有多少个玩具？'),
+                dict(role='BOT', prompt='A: 她有5个玩具。他从妈妈那里得到了2个，所以之后他有5 + 2 = 7个玩具。然后他从爸爸那里得到了2个，所以总共他有7 + 2 = 9个玩具。答案是 9\n'),
+                dict(role='HUMAN', prompt='Q: 服务器房间里有九台电脑。从周一到周四每天增加五台电脑。现在服务器房里有多少台电脑？'),
+                dict(role='BOT', prompt='A: 从周一到周四有4天。每天增加5台电脑。这意味着总共增加了4 * 5 = 20台电脑。一开始有9台电脑，所以现在有9 + 20 = 29台电脑。答案是 29\n'),
+                dict(role='HUMAN', prompt='Q: 迈克尔有58个高尔夫球。星期二，他丢失了23个高尔夫球。星期三，他又丢失了2个。星期三结束时他还剩下多少个高尔夫球？'),
+                dict(role='BOT', prompt='A: 迈克尔一开始有58个球。星期二他丢失了23个，所以之后他还剩下58 - 23 = 35个球。星期三他又丢失了2个，所以现在他还剩下35 - 2 = 33个球。答案是 33\n'),
+                dict(role='HUMAN', prompt='Q: 奥利弗有23美元。她用每个3美元的价格买了五个百吉饼。她还剩下多少钱？'),
+                dict(role='BOT', prompt='A: 她以每个3美元的价格买了5个百吉饼。这意味着她在百吉饼上花费了5 * 3 = 15美元。她一开始有23美元，所以现在她还剩下23 - 15 = 8美元。答案是 8\n'),
+                dict(role='HUMAN', prompt='Q: {question}'),
+                dict(role='BOT', prompt='A: {answer}'),
+                ],
+    "cloze_en": [
+                dict(role='HUMAN', prompt='Q: There are 15 trees in the grove. Grove workers will plant trees in the grove today. After they are done, there will be 21 trees. How many trees did the grove workers plant today?'),
+                dict(role='BOT', prompt='A: We start with 15 trees. Later we have 21 trees. The difference must be the number of trees they planted. So, they must have planted 21 - 15 = 6 trees. The answer is 6.\n'),
+                dict(role='HUMAN', prompt='Q: If there are 3 cars in the parking lot and 2 more cars arrive, how many cars are in the parking lot?'),
+                dict(role='BOT', prompt='A: There are 3 cars in the parking lot already. 2 more arrive. Now there are 3 + 2 = 5 cars. The answer is 5.\n'),
+                dict(role='HUMAN', prompt='Q: Leah had 32 chocolates and her sister had 42. If they ate 35, how many pieces do they have left in total?'),
+                dict(role='BOT', prompt="A: Leah had 32 chocolates and Leah's sister had 42. That means there were originally 32 + 42 = 74 chocolates. 35 have been eaten. So in total they still have 74 - 35 = 39 chocolates. The answer is 39.\n"),
+                dict(role='HUMAN', prompt='Q: Jason had 20 lollipops. He gave Denny some lollipops. Now Jason has 12 lollipops. How many lollipops did Jason give to Denny?'),
+                dict(role='BOT', prompt='A: Jason had 20 lollipops. Since he only has 12 now, he must have given the rest to Denny. The number of lollipops he has given to Denny must have been 20 - 12 = 8 lollipops. The answer is 8.\n'),
+                dict(role='HUMAN', prompt='Q: Shawn has five toys. For Christmas, he got two toys each from his mom and dad. How many toys does he have now?'),
+                dict(role='BOT', prompt='A: He has 5 toys. He got 2 from mom, so after that he has 5 + 2 = 7 toys. Then he got 2 more from dad, so in total he has 7 + 2 = 9 toys. The answer is 9.\n'),
+                dict(role='HUMAN', prompt='Q: There were nine computers in the server room. Five more computers were installed each day, from monday to thursday. How many computers are now in the server room?'),
+                dict(role='BOT', prompt='A: There are 4 days from monday to thursday. 5 computers were added each day. That means in total 4 * 5 = 20 computers were added. There were 9 computers in the beginning, so now there are 9 + 20 = 29 computers. The answer is 29.\n'),
+                dict(role='HUMAN', prompt='Q: Michael had 58 golf balls. On tuesday, he lost 23 golf balls. On wednesday, he lost 2 more. How many golf balls did he have at the end of wednesday?'),
+                dict(role='BOT', prompt='A: Michael initially had 58 balls. He lost 23 on Tuesday, so after that he has 58 - 23 = 35 balls. On Wednesday he lost 2 more so now he has 35 - 2 = 33 balls. The answer is 33.\n'),
+                dict(role='HUMAN', prompt='Q: Olivia has $23. She bought five bagels for $3 each. How much money does she have left?'),
+                dict(role='BOT', prompt='A: She bought 5 bagels for $3 each. This means she spent 5 * $3 = $15 on the bagels. She had $23 in beginning, so now she has $23 - $15 = $8. The answer is 8.\n'),
+                dict(role='HUMAN', prompt='Q: {question}'),
+                dict(role='BOT', prompt='A: {answer}\n'),
+]}
+
+mathbench_sets = {
+    'college': ['single_choice_cn', 'single_choice_en'],
+    'high': ['single_choice_cn', 'single_choice_en'],
+    'middle': ['single_choice_cn', 'single_choice_en'],
+    'primary': ['cloze_cn', 'cloze_en'],
+    'calculate': ['cloze_en'],
+}
+
+# Generate reasoning path or not, only for single choice
+with_reasoning = True
+
+# Use circular evaluation or not
+with_circular_eval = True
+
+mathbench_datasets = []
+
+for _split in list(mathbench_sets.keys()):
+    for _name in mathbench_sets[_split]:
+        mathbench_infer_cfg = dict(
+            prompt_template=dict(
+                type=PromptTemplate,
+                template=dict(
+                    round=[
+                        dict(
+                            role="HUMAN",
+                            prompt=single_choice_prompts[_name + "_with_reasoning"] if with_reasoning else single_choice_prompts[_name],
+                        ),
+                        dict(role="BOT", prompt="{answer}")] if 'choice' in _name else cloze_prompts[_name],
+                    ),
+            ),
+            retriever=dict(type=ZeroRetriever),
+            inferencer=dict(type=GenInferencer, max_out_len=512),
+        )
+
+        mathbench_eval_cfg = dict(
+            evaluator=dict(type=CircularEvaluator if 'choice' in _name and with_circular_eval else AccEvaluator),
+            pred_postprocessor=dict(type=first_option_postprocess, options='ABCD') if 'single_choice' in _name else dict(type=mathbench_postprocess, name=_name))
+
+        mathbench_datasets.append(
+            dict(
+                abbr="mathbench-" + _split + '-' + _name,
+                type=MathBenchDataset,
+                path=f"./data/mathbench/{_split}",
+                name=_name,
+                with_circular=with_circular_eval,
+                reader_cfg=dict(
+                    input_columns=["question"],
+                    output_column="answer"
+                    ),
+                infer_cfg=mathbench_infer_cfg,
+                eval_cfg=mathbench_eval_cfg,
+            ))

configs/datasets/MathBench/mathbench_gen.py

 from mmengine.config import read_base
 
 with read_base():
-    from .mathbench_gen_7b734b import mathbench_datasets  # noqa: F401, F403
+    from .mathbench_2024_gen_de9ff9 import mathbench_datasets  # noqa: F401, F403

configs/datasets/contamination/ceval_contamination_ppl_810ec6.py

+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import PPLOnlyInferencer
+from opencompass.openicl.icl_evaluator import AveragePPLEvaluator
+from opencompass.datasets import JsonlDataset
+
+ceval_datasets = []
+
+ceval_infer_cfg = dict(
+    prompt_template=dict(type=PromptTemplate, template="{text}"),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=PPLOnlyInferencer),
+)
+
+ceval_eval_cfg = dict(evaluator=dict(type=AveragePPLEvaluator))
+
+ceval_reader_cfg = dict(
+    input_columns=['text'],
+    output_column=None,
+)
+
+ceval_datasets.append(
+    dict(
+        abbr=f'ceval-val-ppl',
+        type=JsonlDataset,
+        path='/mnt/petrelfs/zhoufengzhe/repos/cscripts/mock-datas/ceval_val_content.jsonl',
+        reader_cfg=ceval_reader_cfg,
+        infer_cfg=ceval_infer_cfg,
+        eval_cfg=ceval_eval_cfg
+    )
+)
+
+ceval_infer_cfg = dict(
+    prompt_template=dict(type=PromptTemplate, template="{rephrase}"),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=PPLOnlyInferencer),
+)
+
+ceval_eval_cfg = dict(evaluator=dict(type=AveragePPLEvaluator))
+
+ceval_reader_cfg = dict(
+    input_columns=['rephrase'],
+    output_column=None,
+)
+
+ceval_datasets.append(
+    dict(
+        abbr=f'ceval-ref-ppl',
+        type=JsonlDataset,
+        path='/mnt/petrelfs/zhoufengzhe/repos/cscripts/mock-datas/ceval_val_content.jsonl',
+        reader_cfg=ceval_reader_cfg,
+        infer_cfg=ceval_infer_cfg,
+        eval_cfg=ceval_eval_cfg
+    )
+)

configs/datasets/contamination/mbpp_contamination_ppl_f01cb6.py

+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import PPLOnlyInferencer
+from opencompass.openicl.icl_evaluator import AveragePPLEvaluator
+from opencompass.datasets import SanitizedMBPPDataset, JsonlDataset
+
+mbpp_datasets = []
+
+mbpp_infer_cfg = dict(
+    prompt_template=dict(type=PromptTemplate, template="{text}\n{code}"),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=PPLOnlyInferencer),
+)
+
+mbpp_eval_cfg = dict(evaluator=dict(type=AveragePPLEvaluator))
+
+for split in ['train', 'test']:
+    mbpp_reader_cfg = dict(
+        input_columns=['text', 'code'],
+        output_column=None,
+        train_split=split,
+        test_split=split,
+    )
+    mbpp_datasets.append(
+        dict(
+            abbr=f'mbpp-{split}-ppl',
+            type=SanitizedMBPPDataset,
+            path='./data/mbpp/sanitized-mbpp.jsonl',
+            reader_cfg=mbpp_reader_cfg,
+            infer_cfg=mbpp_infer_cfg,
+            eval_cfg=mbpp_eval_cfg)
+    )
+
+
+mbpp_infer_cfg = dict(
+    prompt_template=dict(type=PromptTemplate, template="{text}"),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=PPLOnlyInferencer),
+)
+
+mbpp_eval_cfg = dict(evaluator=dict(type=AveragePPLEvaluator))
+
+mbpp_reader_cfg = dict(
+    input_columns=['text'],
+    output_column=None,
+)
+
+mbpp_datasets.append(
+    dict(
+        abbr=f'mbpp-ref-ppl',
+        type=JsonlDataset,
+        path='/mnt/petrelfs/zhoufengzhe/repos/cscripts/mock-datas/mock_mbpp_20240113.jsonl',
+        reader_cfg=mbpp_reader_cfg,
+        infer_cfg=mbpp_infer_cfg,
+        eval_cfg=mbpp_eval_cfg
+    )
+)

configs/datasets/contamination/mmlu_contamination_ppl_810ec6.py

+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import PPLOnlyInferencer
+from opencompass.openicl.icl_evaluator import AveragePPLEvaluator
+from opencompass.datasets import JsonlDataset
+
+mmlu_datasets = []
+
+mmlu_infer_cfg = dict(
+    prompt_template=dict(type=PromptTemplate, template="{text}"),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=PPLOnlyInferencer),
+)
+
+mmlu_eval_cfg = dict(evaluator=dict(type=AveragePPLEvaluator))
+
+mmlu_reader_cfg = dict(
+    input_columns=['text'],
+    output_column=None,
+)
+
+mmlu_datasets.append(
+    dict(
+        abbr=f'mmlu-test-ppl',
+        type=JsonlDataset,
+        path='/mnt/petrelfs/zhoufengzhe/repos/cscripts/mock-datas/mmlu_test_content.jsonl',
+        reader_cfg=mmlu_reader_cfg,
+        infer_cfg=mmlu_infer_cfg,
+        eval_cfg=mmlu_eval_cfg
+    )
+)
+
+mmlu_infer_cfg = dict(
+    prompt_template=dict(type=PromptTemplate, template="{rephrase}"),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=PPLOnlyInferencer),
+)
+
+mmlu_eval_cfg = dict(evaluator=dict(type=AveragePPLEvaluator))
+
+mmlu_reader_cfg = dict(
+    input_columns=['rephrase'],
+    output_column=None,
+)
+
+mmlu_datasets.append(
+    dict(
+        abbr=f'mmlu-ref-ppl',
+        type=JsonlDataset,
+        path='/mnt/petrelfs/zhoufengzhe/repos/cscripts/mock-datas/mmlu_test_content.jsonl',
+        reader_cfg=mmlu_reader_cfg,
+        infer_cfg=mmlu_infer_cfg,
+        eval_cfg=mmlu_eval_cfg
+    )
+)

configs/datasets/humaneval/humaneval_gen_a82cae.py

 from opencompass.openicl.icl_prompt_template import PromptTemplate
 from opencompass.openicl.icl_retriever import ZeroRetriever
 from opencompass.openicl.icl_inferencer import GenInferencer
-from opencompass.datasets import HumanevalDataset, HumanEvaluator, humaneval_postprocess
+from opencompass.datasets import HumanevalDataset, HumanEvaluator, humaneval_postprocess_v2
 
 humaneval_reader_cfg = dict(
     input_columns=['prompt'], output_column='task_id', train_split='test')
@@ -22,7 +22,7 @@ humaneval_eval_cfg = dict(
     evaluator=dict(type=HumanEvaluator),
     pred_role='BOT',
     k=[1, 10, 100],  # the parameter only for humaneval
-    pred_postprocessor=dict(type=humaneval_postprocess),
+    pred_postprocessor=dict(type=humaneval_postprocess_v2),
 )
 
 humaneval_datasets = [

configs/datasets/mmlu/mmlu_gen_a484b3.py

@@ -3,7 +3,7 @@ from opencompass.openicl.icl_retriever import FixKRetriever
 from opencompass.openicl.icl_inferencer import GenInferencer
 from opencompass.openicl.icl_evaluator import AccEvaluator
 from opencompass.datasets import MMLUDataset
-from opencompass.utils.text_postprocessors import first_capital_postprocess
+from opencompass.utils.text_postprocessors import first_option_postprocess
 
 # None of the mmlu dataset in huggingface is correctly parsed, so we use our own dataset reader
 # Please download the dataset from https://people.eecs.berkeley.edu/~hendrycks/data.tar
@@ -108,7 +108,7 @@ for _name in mmlu_all_sets:
 
     mmlu_eval_cfg = dict(
         evaluator=dict(type=AccEvaluator),
-        pred_postprocessor=dict(type=first_capital_postprocess))
+        pred_postprocessor=dict(type=first_option_postprocess, options='ABCD'))
 
     mmlu_datasets.append(
         dict(

configs/datasets/siqa/siqa_gen_18632c.py

 from opencompass.openicl.icl_prompt_template import PromptTemplate
 from opencompass.openicl.icl_retriever import ZeroRetriever
 from opencompass.openicl.icl_inferencer import GenInferencer
-from opencompass.datasets.custom import OptionSimAccEvaluator
+from opencompass.openicl.icl_evaluator import AccEvaluator
+from opencompass.utils.text_postprocessors import first_option_postprocess
 from opencompass.datasets import siqaDataset_V3
 
 siqa_reader_cfg = dict(
@@ -26,8 +27,8 @@ siqa_infer_cfg = dict(
 )
 
 siqa_eval_cfg = dict(
-    evaluator=dict(type=OptionSimAccEvaluator, options='ABC'),
-    pred_role="BOT",
+            evaluator=dict(type=AccEvaluator),
+            pred_postprocessor=dict(type=first_option_postprocess, options='ABC')
 )
 
 siqa_datasets = [

configs/datasets/teval/README.md

+# T-Eval
+
+Tool utilization is comprehensively decomposed into multiple sub-processes, including instruction following, planning, reasoning, retrieval, understanding, and review. Based on that, T-Eval is introduced to evaluate the tool-utilization capability step by step. T-Eval disentangles the tool utilization evaluation into several sub-domains along model capabilities, facilitating the inner understanding of both holistic and isolated competency of LLMs.
+
+[Paper](https://arxiv.org/abs/2312.14033)
+
+[Project Page](https://open-compass.github.io/T-Eval/)
+
+[LeaderBoard](https://open-compass.github.io/T-Eval/leaderboard.html)
+
+[HuggingFace](https://huggingface.co/datasets/lovesnowbest/T-Eval)
+
+## Citation
+
+```
+@article{chen2023t,
+  title={T-Eval: Evaluating the Tool Utilization Capability Step by Step},
+  author={Chen, Zehui and Du, Weihua and Zhang, Wenwei and Liu, Kuikun and Liu, Jiangning and Zheng, Miao and Zhuo, Jingming and Zhang, Songyang and Lin, Dahua and Chen, Kai and others},
+  journal={arXiv preprint arXiv:2312.14033},
+  year={2023}
+}
+```

configs/datasets/teval/teval_en_gen.py

+from mmengine.config import read_base
+
+with read_base():
+    from .teval_en_gen_1ac254 import teval_datasets

configs/datasets/teval/teval_en_gen_1ac254.py

+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import ChatInferencer
+from opencompass.openicl.icl_evaluator import TEvalEvaluator
+from opencompass.datasets import teval_postprocess, TEvalDataset
+
+teval_subject_mapping = {
+    "instruct": ["instruct_v1"],
+    "plan": ["plan_json_v1", "plan_str_v1"],
+    "review": ["review_str_v1"],
+    "reason_retrieve_understand": ["reason_retrieve_understand_json_v1"],
+    "reason": ["reason_str_v1"],
+    "retrieve": ["retrieve_str_v1"],
+    "understand": ["understand_str_v1"],
+}
+
+teval_reader_cfg = dict(input_columns=["prompt"], output_column="ground_truth")
+
+teval_infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template=dict(
+            round=[
+                dict(role="HUMAN", prompt="{prompt}"),
+            ],
+        ),
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=ChatInferencer),
+)
+
+teval_all_sets = list(teval_subject_mapping.keys())
+
+teval_datasets = []
+for _name in teval_all_sets:
+    teval_eval_cfg = dict(
+        evaluator=dict(type=TEvalEvaluator, subset=_name),
+        pred_postprocessor=dict(type=teval_postprocess),
+        num_gpus=1,
+    )
+    for subset in teval_subject_mapping[_name]:
+        teval_datasets.append(
+            dict(
+                abbr="teval-" + subset,
+                type=TEvalDataset,
+                path="./data/teval/EN",
+                name=subset,
+                reader_cfg=teval_reader_cfg,
+                infer_cfg=teval_infer_cfg,
+                eval_cfg=teval_eval_cfg,
+            )
+        )

configs/datasets/teval/teval_zh_gen.py

+from mmengine.config import read_base
+
+with read_base():
+    from .teval_zh_gen_1ac254 import teval_datasets

configs/datasets/teval/teval_zh_gen_1ac254.py

+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import ChatInferencer
+from opencompass.openicl.icl_evaluator import TEvalEvaluator
+from opencompass.datasets import teval_postprocess, TEvalDataset
+
+teval_subject_mapping = {
+    "instruct_zh": ["instruct_v1_zh"],
+    "plan_zh": ["plan_json_v1_zh", "plan_str_v1_zh"],
+    "review_zh": ["review_str_v1_zh"],
+    "reason_retrieve_understand_zh": ["reason_retrieve_understand_json_v1_zh"],
+    "reason_zh": ["reason_str_v1_zh"],
+    "retrieve_zh": ["retrieve_str_v1_zh"],
+    "understand_zh": ["understand_str_v1_zh"],
+}
+
+teval_reader_cfg = dict(input_columns=["prompt"], output_column="ground_truth")
+
+teval_infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template=dict(
+            round=[
+                dict(role="HUMAN", prompt="{prompt}"),
+            ],
+        ),
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=ChatInferencer),
+)
+
+teval_all_sets = list(teval_subject_mapping.keys())
+
+teval_datasets = []
+for _name in teval_all_sets:
+    teval_eval_cfg = dict(
+        evaluator=dict(type=TEvalEvaluator, subset=_name),
+        pred_postprocessor=dict(type=teval_postprocess),
+        num_gpus=1,
+    )
+    for subset in teval_subject_mapping[_name]:
+        teval_datasets.append(
+            dict(
+                abbr="teval-" + subset,
+                type=TEvalDataset,
+                path="./data/teval/ZH",
+                name=subset,
+                reader_cfg=teval_reader_cfg,
+                infer_cfg=teval_infer_cfg,
+                eval_cfg=teval_eval_cfg,
+            )
+        )

configs/eval_internlm_math_chat.py

+from mmengine.config import read_base
+from opencompass.models.huggingface import HuggingFaceCausalLM
+
+with read_base():
+    # choose a list of datasets
+    from .datasets.gsm8k.gsm8k_gen import gsm8k_datasets
+    from .datasets.math.math_gen_736506 import math_datasets
+
+    from .models.hf_internlm.hf_internlm2_chat_math_7b import models as internlm_math_chat_7b_models
+    from .models.hf_internlm.hf_internlm2_chat_math_20b import models as internlm_math_chat_20b_models
+
+datasets = sum((v for k, v in locals().items() if k.endswith('_datasets')), [])
+# Eval Math and GSM8k for both Internlm-Math-Chat-7B and 20b
+datasets = [*math_datasets, *gsm8k_datasets]
+models = [*internlm_math_chat_7b_models, *internlm_math_chat_20b_models]

configs/eval_teval.py

+from copy import deepcopy
+from mmengine.config import read_base
+
+with read_base():
+    from .datasets.teval.teval_en_gen_1ac254 import teval_datasets as teval_en_datasets
+    from .datasets.teval.teval_zh_gen_1ac254 import teval_datasets as teval_zh_datasets
+
+    from .models.qwen.hf_qwen_7b_chat import models as hf_qwen_7b_chat_model
+    from .models.hf_internlm.hf_internlm2_chat_7b import models as hf_internlm2_chat_7b_model
+    from .models.hf_llama.hf_llama2_7b_chat import models as hf_llama2_7b_chat_model
+
+    from .summarizers.teval import summarizer
+
+meta_template_system_patches = {
+    'internlm2-chat-7b-hf': dict(role='SYSTEM', begin='<|im_start|>system\n', end='<|im_end|>\n'),
+    'internlm2-chat-20b-hf': dict(role='SYSTEM', begin='<|im_start|>system\n', end='<|im_end|>\n'),
+}
+
+_origin_models = sum([v for k, v in locals().items() if k.endswith("_model")], [])
+models = []
+for m in _origin_models:
+    m = deepcopy(m)
+    if 'meta_template' in m and 'round' in m['meta_template']:
+        round = m['meta_template']['round']
+        if all(r['role'].upper() != 'SYSTEM' for r in round):  # no system round
+            if m['abbr'] in meta_template_system_patches:
+                system_round = meta_template_system_patches[m['abbr']]
+            else:
+                system_round = [r for r in round if r['role'].upper() == 'HUMAN'][0]
+                system_round = deepcopy(system_round)
+                system_round['role'] = 'SYSTEM'
+            m['meta_template']['round'].append(system_round)
+    else:
+        raise ValueError(f'no meta_template.round in {m.get("abbr", None)}')
+
+    print(f'model {m["abbr"]} is using the following meta_template: {m["meta_template"]}')
+    models.append(m)
+
+datasets = teval_en_datasets + teval_zh_datasets
+work_dir = './outputs/teval'
+
+
+'''
+dataset                                      version    metric          mode       qwen-7b-chat-hf    internlm2-chat-7b-hf    llama-2-7b-chat-hf
+-------------------------------------------  ---------  --------------  -------  -----------------  ----------------------  --------------------
+teval                                        -          naive_average   unknown              57.69                   78.18                 36.63
+teval-instruct_v1                            10482d     string_metric   unknown              28.83                   98.08                 50.27
+teval-instruct_v1                            10482d     json_metric     unknown              94.32                   97.08                  0.15
+teval-plan_str_v1                            10482d     f1_score        unknown              66.24                   84.12                 45.72
+teval-plan_json_v1                           10482d     f1_score        unknown              63.62                   77.71                 19.95
+teval-reason_str_v1                          10482d     thought         unknown              54.14                   63.58                 44.92
+teval-reason_retrieve_understand_json_v1     10482d     thought         unknown              33.77                   54.72                 21.49
+teval-retrieve_str_v1                        10482d     name            unknown              73.89                   85.28                 60.6
+teval-reason_retrieve_understand_json_v1     10482d     name            unknown              31.15                   68.97                 15.34
+teval-understand_str_v1                      10482d     args            unknown              77.76                   93.03                 65.61
+teval-reason_retrieve_understand_json_v1     10482d     args            unknown              44.16                   72.23                 26.84
+teval-review_str_v1                          10482d     review_quality  unknown              62.22                   71.66                 44.35
+teval_zh                                     -          naive_average   unknown              61.31                   75.01                 32.33
+teval-instruct_v1_zh                         10482d     string_metric   unknown              88.69                   98.19                 23.64
+teval-instruct_v1_zh                         10482d     json_metric     unknown              75.77                   96.62                  0.89
+teval-plan_str_v1_zh                         10482d     f1_score        unknown              62.43                   70.69                 47.82
+teval-plan_json_v1_zh                        10482d     f1_score        unknown              61.46                   68.95                 15.87
+teval-reason_str_v1_zh                       10482d     thought         unknown              59.43                   68.14                 46.96
+teval-reason_retrieve_understand_json_v1_zh  10482d     thought         unknown              39.19                   60.37                 23.91
+teval-retrieve_str_v1_zh                     10482d     name            unknown              69.41                   84.22                 54.44
+teval-reason_retrieve_understand_json_v1_zh  10482d     name            unknown              32.87                   70.46                 14.16
+teval-understand_str_v1_zh                   10482d     args            unknown              84.39                   88.62                 77.29
+teval-reason_retrieve_understand_json_v1_zh  10482d     args            unknown              48.71                   72.71                 28.83
+teval-review_str_v1_zh                       10482d     review_quality  unknown              56.67                   60.57                 27.1
+'''

configs/models/hf_internlm/hf_internlm2_1_8b.py

+from opencompass.models import HuggingFaceCausalLM
+
+
+models = [
+    dict(
+        type=HuggingFaceCausalLM,
+        abbr='internlm2-1.8b-hf',
+        path="internlm/internlm2-1_8b",
+        tokenizer_path='internlm/internlm2-1_8b',
+        model_kwargs=dict(
+            trust_remote_code=True,
+            device_map='auto',
+        ),
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            use_fast=False,
+            trust_remote_code=True,
+        ),
+        max_out_len=100,
+        min_out_len=1,
+        max_seq_len=2048,
+        batch_size=8,
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]

configs/models/hf_internlm/hf_internlm2_20b.py

@@ -18,6 +18,7 @@ models = [
             trust_remote_code=True,
         ),
         max_out_len=100,
+        min_out_len=1,
         max_seq_len=2048,
         batch_size=8,
         run_cfg=dict(num_gpus=2, num_procs=1),

configs/models/hf_internlm/hf_internlm2_7b.py

@@ -18,7 +18,7 @@ models = [
             trust_remote_code=True,
         ),
         max_out_len=100,
-        min_out_len=3,
+        min_out_len=1,
         max_seq_len=2048,
         batch_size=8,
         run_cfg=dict(num_gpus=1, num_procs=1),