[Datasets] MedQA, ProteinLMBench; Add Models: huatuogpt, baichuanM1 (#2064)

* Add Datasets: MedQA, ProteinLMBench; Add Models: huatuogpt, baichuanM1

* Fix bugs for MedQA. Add info in dataset-index

* Add version code for MedQA and ProteinLMBench

* Add version code for MedQA and ProteinLMBench

dataset-index.yml

@@ -122,6 +122,12 @@
     paper: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10778138
     configpath: opencompass/configs/datasets/MedBench/medbench_gen.py
     configpath_llmjudge: ''
+- MedXpertQA:
+    name: MedQA
+    category: Knowledge / Medicine
+    paper: https://arxiv.org/abs/2009.13081
+    configpath: opencompass/configs/datasets/MedQA/MedQA_gen.py
+    configpath_llmjudge: opencompass/configs/datasets/MedQA/MedQA_llmjudge_gen.py
 - MedXpertQA:
     name: MedXpertQA
     category: Knowledge / Medicine
@@ -763,6 +769,12 @@
     paper: https://arxiv.org/pdf/1911.11641v1
     configpath: opencompass/configs/datasets/piqa/piqa_gen.py
     configpath_llmjudge: ''
+- ProteinLMBench:
+    name: ProteinLMBench
+    category: Knowledge / Biology (Protein)
+    paper: https://arxiv.org/abs/2406.05540
+    configpath: opencompass/configs/datasets/ProteinLMBench/ProteinLMBench_gen.py
+    configpath_llmjudge: opencompass/configs/datasets/ProteinLMBench/ProteinLMBench_llmjudge_gen.py
 - py150:
     name: py150
     category: Code

opencompass/configs/datasets/MedQA/MedQA_gen_3bf756.py

@@ -0,0 +1,63 @@
+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 AccEvaluator
+from opencompass.utils.text_postprocessors import first_option_postprocess
+from opencompass.datasets.MedQA import MedQADataset
+
+
+QUERY_TEMPLATE = """
+Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: $LETTER' (without quotes) where LETTER is one of Options(e.g. one of ABCDEFGHIJKLMNOP). Think step by step before answering.
+
+Question:\n
+{question}
+
+Options:\n
+{choices}
+
+""".strip()
+
+
+MedQA_datasets = []
+
+MedQA_reader_cfg = dict(
+    input_columns=['question', 'choices'],
+    output_column='label',
+)
+
+MedQA_infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template=dict(
+            round=[
+                dict(role='HUMAN', prompt=QUERY_TEMPLATE),
+            ],
+        ),
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=GenInferencer),
+)
+
+MedQA_subsets = {
+    'US': 'xuxuxuxuxu/MedQA_US_test',
+    'Mainland': 'xuxuxuxuxu/MedQA_Mainland_test',
+    'Taiwan': 'xuxuxuxuxu/MedQA_Taiwan_test',
+}
+
+for split in list(MedQA_subsets.keys()):
+
+    MedQA_eval_cfg = dict(
+        evaluator=dict(type=AccEvaluator),
+        pred_postprocessor=dict(type=first_option_postprocess, options='ABCD')
+    )
+
+    MedQA_datasets.append(
+        dict(
+            abbr=f'MedQA_{split}',
+            type=MedQADataset,
+            path=MedQA_subsets[split],
+            reader_cfg=MedQA_reader_cfg,
+            infer_cfg=MedQA_infer_cfg,
+            eval_cfg=MedQA_eval_cfg,
+        )
+    )

opencompass/configs/datasets/MedQA/MedQA_llmjudge_gen_3bf756.py

@@ -0,0 +1,108 @@
+from mmengine.config import read_base
+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import GenInferencer
+from opencompass.evaluator import GenericLLMEvaluator
+from opencompass.datasets import generic_llmjudge_postprocess
+from opencompass.datasets.MedQA import MedQADataset
+
+
+QUERY_TEMPLATE = """
+Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: $LETTER' (without quotes) where LETTER is one of Options(e.g. one of ABCDEFGHIJKLMNOP). Think step by step before answering.
+
+Question:\n
+{question}
+
+Options:\n
+{choices}
+
+""".strip()
+
+GRADER_TEMPLATE = """
+    Please as a grading expert, judge whether the final answers given by the candidates below are consistent with the standard answers, that is, whether the candidates answered correctly. 
+    
+    Here are some evaluation criteria:
+    1. Please refer to the given standard answer. You don't need to re-generate the answer to the question because the standard answer has been given. You only need to judge whether the candidate's answer is consistent with the standard answer according to the form of the question. Don't try to answer the original question. You can assume that the standard answer is definitely correct.
+    2. Because the candidate's answer may be different from the standard answer in the form of expression, before making a judgment, please understand the question and the standard answer first, and then judge whether the candidate's answer is correct, but be careful not to try to answer the original question.
+    3. Some answers may contain multiple items, such as multiple-choice questions, multiple-select questions, fill-in-the-blank questions, etc. As long as the answer is the same as the standard answer, it is enough. For multiple-select questions and multiple-blank fill-in-the-blank questions, the candidate needs to answer all the corresponding options or blanks correctly to be considered correct.
+    4. Some answers may be expressed in different ways, such as some answers may be a mathematical expression, some answers may be a textual description, as long as the meaning expressed is the same. And some formulas are expressed in different ways, but they are equivalent and correct.
+
+    Please judge whether the following answers are consistent with the standard answer based on the above criteria. Grade the predicted answer of this new question as one of:
+    A: CORRECT 
+    B: INCORRECT
+    Just return the letters "A" or "B", with no text around it.
+
+    Here is your task. Simply reply with either CORRECT, INCORRECT. Don't apologize or correct yourself if there was a mistake; we are just trying to grade the answer.
+
+    <Original Question Begin>: {question}\n {choices} \n<Original Question End>\n\n
+    <Gold Target Begin>: \n{label}\n<Gold Target End>\n\n
+    <Predicted Answer Begin>: \n{prediction}\n<Predicted End>\n\n
+    Judging the correctness of candidates' answers:
+""".strip()
+
+MedQA_datasets = []
+
+MedQA_reader_cfg = dict(
+    input_columns=['question', 'choices'],
+    output_column='label',
+)
+
+MedQA_infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template=dict(
+            round=[
+                dict(role='HUMAN', prompt=QUERY_TEMPLATE),
+            ],
+        ),
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=GenInferencer),
+)
+
+MedQA_subsets = {
+    'US': 'xuxuxuxuxu/MedQA_US_test',
+    'Mainland': 'xuxuxuxuxu/MedQA_Mainland_test',
+    'Taiwan': 'xuxuxuxuxu/MedQA_Taiwan_test',
+}
+
+for split in list(MedQA_subsets.keys()):
+
+    MedQA_eval_cfg = dict(
+        evaluator=dict(
+            type=GenericLLMEvaluator,
+            prompt_template=dict(
+                type=PromptTemplate,
+                template=dict(
+                    begin=[
+                        dict(
+                            role='SYSTEM',
+                            fallback_role='HUMAN',
+                            prompt="You are a helpful assistant who evaluates the correctness and quality of models' outputs.",
+                        )
+                    ],
+                    round=[
+                        dict(role='HUMAN', prompt=GRADER_TEMPLATE),
+                    ],
+                ),
+            ),
+            dataset_cfg=dict(
+                type=MedQADataset,
+                path=MedQA_subsets[split],
+                reader_cfg=MedQA_reader_cfg,
+            ),
+            judge_cfg=dict(),
+            dict_postprocessor=dict(type=generic_llmjudge_postprocess),
+        ),
+    )
+
+    MedQA_datasets.append(
+        dict(
+            abbr=f'MedQA_{split}',
+            type=MedQADataset,
+            path=MedQA_subsets[split],
+            reader_cfg=MedQA_reader_cfg,
+            infer_cfg=MedQA_infer_cfg,
+            eval_cfg=MedQA_eval_cfg,
+        )
+    )

opencompass/configs/datasets/ProteinLMBench/ProteinLMBench_gen_a67965.py

@@ -0,0 +1,46 @@
+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 AccEvaluator
+from opencompass.datasets.ProteinLMBench import ProteinLMBenchDataset, ProteinLMBenchEvaluator
+
+QUERY_TEMPLATE = "Answer the following multiple choice question. There is only one correct answer. The last line of your response should be in the format 'Answer: $LETTER' (without quotes), where LETTER is the letter among {start} through {end}.\n{question}"
+
+
+# Reader configuration
+reader_cfg = dict(
+    input_columns=['question', 'start', 'end', 'options'],
+    output_column='label',
+)
+
+# Inference configuration
+infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template=dict(
+            round=[
+                dict(
+                    role='HUMAN',
+                    prompt=QUERY_TEMPLATE
+                )
+            ], ),
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=GenInferencer),
+)
+
+# Evaluation configuration
+eval_cfg = dict(
+    evaluator=dict(type=ProteinLMBenchEvaluator),
+)
+
+proteinlmbench_dataset = dict(
+    abbr='ProteinLMBench',
+    type=ProteinLMBenchDataset,
+    path='tsynbio/ProteinLMBench',
+    reader_cfg=reader_cfg,
+    infer_cfg=infer_cfg,
+    eval_cfg=eval_cfg
+)
+
+proteinlmbench_datasets = [proteinlmbench_dataset]

opencompass/configs/datasets/ProteinLMBench/ProteinLMBench_llmjudge_gen_a67965.py

@@ -0,0 +1,89 @@
+from mmengine.config import read_base
+from opencompass.openicl.icl_prompt_template import PromptTemplate
+from opencompass.openicl.icl_retriever import ZeroRetriever
+from opencompass.openicl.icl_inferencer import GenInferencer
+from opencompass.evaluator import GenericLLMEvaluator
+from opencompass.datasets import generic_llmjudge_postprocess
+from opencompass.datasets.ProteinLMBench import ProteinLMBenchDataset
+
+QUERY_TEMPLATE = "Answer the following multiple choice question. There is only one correct answer. The last line of your response should be in the format 'Answer: $LETTER' (without quotes), where LETTER is the letter among {start} through {end}.\n{question}"
+
+GRADER_TEMPLATE = """
+    Please as a grading expert, judge whether the final answers given by the candidates below are consistent with the standard answers, that is, whether the candidates answered correctly. 
+    
+    Here are some evaluation criteria:
+    1. Please refer to the given standard answer. You don't need to re-generate the answer to the question because the standard answer has been given. You only need to judge whether the candidate's answer is consistent with the standard answer according to the form of the question. Don't try to answer the original question. You can assume that the standard answer is definitely correct.
+    2. Because the candidate's answer may be different from the standard answer in the form of expression, before making a judgment, please understand the question and the standard answer first, and then judge whether the candidate's answer is correct, but be careful not to try to answer the original question.
+    3. Some answers may contain multiple items, such as multiple-choice questions, multiple-select questions, fill-in-the-blank questions, etc. As long as the answer is the same as the standard answer, it is enough. For multiple-select questions and multiple-blank fill-in-the-blank questions, the candidate needs to answer all the corresponding options or blanks correctly to be considered correct.
+    4. Some answers may be expressed in different ways, such as some answers may be a mathematical expression, some answers may be a textual description, as long as the meaning expressed is the same. And some formulas are expressed in different ways, but they are equivalent and correct.
+
+    Please judge whether the following answers are consistent with the standard answer based on the above criteria. Grade the predicted answer of this new question as one of:
+    A: CORRECT 
+    B: INCORRECT
+    Just return the letters "A" or "B", with no text around it.
+
+    Here is your task. Simply reply with either CORRECT, INCORRECT. Don't apologize or correct yourself if there was a mistake; we are just trying to grade the answer.
+
+    <Original Question Begin>: {question}\n<Original Question End>\n\n
+    <Gold Target Begin>: \n{label}\n<Gold Target End>\n\n
+    <Predicted Answer Begin>: \n{prediction}\n<Predicted End>\n\n
+    Judging the correctness of candidates' answers:
+""".strip()
+
+
+reader_cfg = dict(
+    input_columns=['question', 'start', 'end', 'options'],
+    output_column='label',
+)
+
+infer_cfg = dict(
+    prompt_template=dict(
+        type=PromptTemplate,
+        template=dict(
+            round=[
+                dict(role='HUMAN', prompt=QUERY_TEMPLATE),
+            ],
+        ),
+    ),
+    retriever=dict(type=ZeroRetriever),
+    inferencer=dict(type=GenInferencer),
+)
+
+eval_cfg = dict(
+    evaluator=dict(
+        type=GenericLLMEvaluator,
+        prompt_template=dict(
+            type=PromptTemplate,
+            template=dict(
+                begin=[
+                    dict(
+                        role='SYSTEM',
+                        fallback_role='HUMAN',
+                        prompt="You are a helpful assistant who evaluates the correctness and quality of models' outputs.",
+                    )
+                ],
+                round=[
+                    dict(role='HUMAN', prompt=GRADER_TEMPLATE),
+                ],
+            ),
+        ),
+        dataset_cfg=dict(
+            type=ProteinLMBenchDataset,
+            path='tsynbio/ProteinLMBench',
+            reader_cfg=reader_cfg,
+        ),
+        judge_cfg=dict(),
+        dict_postprocessor=dict(type=generic_llmjudge_postprocess),
+    ),
+)
+
+proteinlmbench_dataset = dict(
+    abbr='ProteinLMBench',
+    type=ProteinLMBenchDataset,
+    path='tsynbio/ProteinLMBench',
+    reader_cfg=reader_cfg,
+    infer_cfg=infer_cfg,
+    eval_cfg=eval_cfg
+)
+
+proteinlmbench_datasets = [proteinlmbench_dataset]

opencompass/configs/models/baichuan/hf_baichuan_m1_14b_base.py

@@ -0,0 +1,14 @@
+import torch
+from opencompass.models import HuggingFaceBaseModel
+
+models = [
+    dict(
+        type=HuggingFaceBaseModel,
+        abbr='baichuan-m1-14b-base-hf',
+        path='baichuan-inc/Baichuan-M1-14B-Base',
+        max_out_len=1024,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto', trust_remote_code=True, torch_dtype=torch.bfloat16),
+        run_cfg=dict(num_gpus=1),
+    )
+]

opencompass/configs/models/baichuan/hf_baichuan_m1_14b_instruct.py

@@ -0,0 +1,14 @@
+import torch
+from opencompass.models import HuggingFacewithChatTemplate
+
+models = [
+    dict(
+        type=HuggingFacewithChatTemplate,
+        abbr='baichuan-m1-14b-instruct-hf',
+        path='baichuan-inc/Baichuan-M1-14B-Instruct',
+        max_out_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto', trust_remote_code=True, torch_dtype=torch.bfloat16),
+        run_cfg=dict(num_gpus=1),
+    )
+]

opencompass/configs/models/huatuogpt/hf_huatuogpt2_13b.py

@@ -0,0 +1,17 @@
+from opencompass.models import HuggingFacewithChatTemplate
+
+models = [
+    dict(
+        type=HuggingFacewithChatTemplate,
+        abbr='huatuogpt2-13b-hf',
+        path='FreedomIntelligence/HuatuoGPT2-13B',
+        tokenizer_kwargs=dict(padding_side='left',
+                              truncation_side='left',
+                              trust_remote_code=True,
+                              use_fast=True,),
+        max_out_len=1024,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto', trust_remote_code=True),
+        run_cfg=dict(num_gpus=4),
+    )
+]

opencompass/configs/models/huatuogpt/hf_huatuogpt2_7b.py

@@ -0,0 +1,13 @@
+from opencompass.models import HuggingFacewithChatTemplate
+
+models = [
+    dict(
+        type=HuggingFacewithChatTemplate,
+        abbr='huatuogpt2-7b-hf',
+        path='FreedomIntelligence/HuatuoGPT2-7B',
+        max_out_len=1024,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto', trust_remote_code=True),
+        run_cfg=dict(num_gpus=1),
+    )
+]

opencompass/configs/models/huatuogpt/hf_huatuogpt_o1_7b.py

@@ -0,0 +1,15 @@
+from opencompass.models import HuggingFacewithChatTemplate
+from opencompass.utils.text_postprocessors import extract_non_reasoning_content
+
+models = [
+    dict(
+        type=HuggingFacewithChatTemplate,
+        abbr='huatuogpt-o1-7b-hf',
+        path='FreedomIntelligence/HuatuoGPT-o1-7B',
+        max_out_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto', trust_remote_code=True),
+        run_cfg=dict(num_gpus=1),
+        pred_postprocessor=dict(type=extract_non_reasoning_content, think_start_token='## Thinking', think_end_token='## Final Response'),
+    )
+]

opencompass/configs/models/huatuogpt/hf_huatuogpt_o1_8b.py

@@ -0,0 +1,15 @@
+from opencompass.models import HuggingFacewithChatTemplate
+from opencompass.utils.text_postprocessors import extract_non_reasoning_content
+
+models = [
+    dict(
+        type=HuggingFacewithChatTemplate,
+        abbr='huatuogpt-o1-8b-hf',
+        path='FreedomIntelligence/HuatuoGPT-o1-8B',
+        max_out_len=2048,
+        batch_size=8,
+        model_kwargs=dict(device_map='auto', trust_remote_code=True),
+        run_cfg=dict(num_gpus=1),
+        pred_postprocessor=dict(type=extract_non_reasoning_content, think_start_token='## Thinking', think_end_token='## Final Response'),
+    )
+]

opencompass/datasets/MedQA.py

@@ -0,0 +1,29 @@
+from datasets import Dataset, load_dataset
+
+from opencompass.registry import LOAD_DATASET
+
+from .base import BaseDataset
+
+
+@LOAD_DATASET.register_module()
+class MedQADataset(BaseDataset):
+
+    @staticmethod
+    def load_single(path):
+        dataset = []
+        ds = load_dataset(path)
+        for data in ds['train']:
+            data['label'] = data['answer_idx']
+            choices = ''
+            for option in data['options']:
+                choices += option + '. ' + data['options'][option] + '\n'
+            data['choices'] = choices
+
+            dataset.append(data)
+
+        return Dataset.from_list(dataset)
+
+    @staticmethod
+    def load(path):
+        dataset = MedQADataset.load_single(path)
+        return dataset

opencompass/datasets/ProteinLMBench.py

@@ -0,0 +1,58 @@
+from datasets import load_dataset
+
+from opencompass.openicl import BaseEvaluator
+from opencompass.registry import LOAD_DATASET
+from opencompass.utils.text_postprocessors import first_option_postprocess
+
+from .base import BaseDataset
+
+
+def _parse(item):
+    item['start'] = chr(65)
+    item['end'] = chr(65 + len(item.get('options', [])) - 1)
+    new_options = []
+    choices = ''
+    for i in range(len(item['options'])):
+        new_options.append(item['options'][i].split(': ')[-1])
+        choices += chr(65 +
+                       i) + '. ' + item['options'][i].split(': ')[-1] + '\n'
+    item['question'] = (f'\nQuestion: {item["question"]}\n'
+                        f'Answer Choices: \n{choices}')
+    item['options'] = new_options
+    item['label'] = chr(65 + int(item['answer'].split(' ')[-1]) -
+                        1)  # Index from 1 in answer
+    return item
+
+
+@LOAD_DATASET.register_module()
+class ProteinLMBenchDataset(BaseDataset):
+
+    @staticmethod
+    def load(path: str, **kwargs):
+        dataset = load_dataset(path, 'evaluation', split='train')
+        dataset = dataset.map(lambda item: _parse(item))
+
+        return dataset
+
+
+class ProteinLMBenchEvaluator(BaseEvaluator):
+
+    def score(self, predictions, references, test_set):
+        if len(predictions) != len(references):
+            return {'error': 'preds and refrs have different length'}
+        correct = 0
+        count = 0
+        details = []
+        for idx, (prediction,
+                  reference) in enumerate(zip(predictions, references)):
+            options = ''.join(
+                [chr(65 + i) for i in range(len(test_set['options'][idx]))])
+            predict = first_option_postprocess(prediction, options)
+            detail = {'pred': predict, 'answer': reference, 'correct': False}
+            count += 1
+            if predict == reference:
+                correct += 1
+                detail['correct'] = True
+            details.append(detail)
+        result = {'accuracy': 100 * correct / count, 'details': details}
+        return result

opencompass/datasets/__init__.py

@@ -97,6 +97,7 @@ from .math_intern import *  # noqa: F401, F403
 from .mathbench import *  # noqa: F401, F403
 from .mbpp import *  # noqa: F401, F403
 from .medbench import *  # noqa: F401, F403
+from .MedQA import *  # noqa: F401, F403
 from .MedXpertQA import *  # noqa: F401, F403
 from .mgsm import *  # noqa: F401, F403
 from .mmlu import *  # noqa: F401, F403
@@ -118,6 +119,7 @@ from .OlympiadBench import *  # noqa: F401, F403
 from .OpenFinData import *  # noqa: F401, F403
 from .physics import *  # noqa: F401, F403
 from .piqa import *  # noqa: F401, F403
+from .ProteinLMBench import *  # noqa: F401, F403
 from .py150 import *  # noqa: F401, F403
 from .qasper import *  # noqa: F401, F403
 from .qaspercut import *  # noqa: F401, F403