crfm-helm 0.5.4__py3-none-any.whl → 0.5.5__py3-none-any.whl

helm/benchmark/scenarios/medbullets_scenario.py

@@ -0,0 +1,140 @@
+import os
+import csv
+import sys
+from typing import List
+
+from helm.benchmark.scenarios.scenario import (
+    CORRECT_TAG,
+    TEST_SPLIT,
+    Input,
+    Instance,
+    Output,
+    Reference,
+    Scenario,
+)
+from helm.common.general import ensure_file_downloaded
+
+csv.field_size_limit(sys.maxsize)
+
+
+class MedBulletsScenario(Scenario):
+    """
+    From "Benchmarking Large Language Models on Answering and Explaining Challenging Medical Questions"
+    (Chen et al.), MedBullet is a dataset comprising USMLE Step 2&3 style clinical questions. The dataset
+    is designed to evaluate the performance of LLMs in answering and explaining challenging medical questions,
+    emphasizing the need for explainable AI in medical QA.
+
+    Example from the dataset:
+
+    Question:
+    A 42-year-old woman is enrolled in a randomized controlled trial to study cardiac function in the setting of
+    several different drugs. She is started on verapamil and instructed to exercise at 50% of her VO2 max while
+    several cardiac parameters are being measured. During this experiment, which of the following represents
+    the relative conduction speed through the heart from fastest to slowest?
+
+    A) AV node > ventricles > atria > Purkinje fibers
+    B) Purkinje fibers > ventricles > atria > AV node
+    C) Purkinje fibers > atria > ventricles > AV node
+    D) Purkinje fibers > AV node > ventricles > atria
+
+    Answer:
+    The answer is C. Explanation: The conduction velocity of the structures of the heart is in the following order:
+    Purkinje fibers > atria > ventricles > AV node. A calcium channel blocker such as verapamil would only slow
+    conduction in the AV node.
+
+    @Article{MedBullet,
+    author = {Hanjie Chen and Zhouxiang Fang and Yash Singla and Mark Dredze},
+    title = {Benchmarking Large Language Models on Answering and Explaining Challenging Medical Questions},
+    year = {2023},
+    abstract = {LLMs have demonstrated impressive performance in answering medical questions, such as passing scores
+    on medical licensing examinations. However, medical board exam questions or general clinical questions do not
+    capture the complexity of realistic clinical cases. Moreover, the lack of reference explanations means we cannot
+    easily evaluate the reasoning of model decisions, a crucial component of supporting doctors in making complex
+    medical decisions. To address these challenges, we construct two new datasets: JAMA Clinical Challenge and
+    Medbullets. JAMA Clinical Challenge consists of questions based on challenging clinical cases, while Medbullets
+    comprises USMLE Step 2&3 style clinical questions. Both datasets are structured as multiple-choice
+    question-answering tasks, where each question is accompanied by an expert-written explanation. We evaluate four
+    LLMs on the two datasets using various prompts. Experiments demonstrate that our datasets are harder than
+    previous benchmarks. The inconsistency between automatic and human evaluations of model-generated explanations
+    highlights the need to develop new metrics to support future research on explainable medical QA.}}
+
+    Task:
+    Given a clinical question with multiple-choice options, models must identify the correct answer and generate a
+    response that includes the reasoning, as described in the expert-written explanation.
+    """
+
+    DATASET_DOWNLOAD_BASE_URL = (
+        "https://raw.githubusercontent.com/HanjieChen/ChallengeClinicalQA/refs/heads/main/medbullets/"
+    )
+
+    name = "medbullet"
+    description = "A USMLE-style medical question dataset with multiple-choice answers and explanations."
+    tags = ["reasoning", "biomedical"]
+
+    # Define the possible answer choices
+    POSSIBLE_ANSWER_CHOICES: List[str] = ["A", "B", "C", "D", "E"]
+
+    def __init__(self):
+        super().__init__()
+        # self.splits = {"_op4": TRAIN_SPLIT, "_op5": TEST_SPLIT}
+        # limit to zero shot setting for now
+        self.splits = {"_op5": TEST_SPLIT}
+
+    def download_csv(self, output_path: str, split: str):
+        """Download CSV files for the given split."""
+        csv_path = os.path.join(output_path, f"medbullets{split}.csv")
+        ensure_file_downloaded(
+            source_url=f"{self.DATASET_DOWNLOAD_BASE_URL}/medbullets{split}.csv",
+            target_path=csv_path,
+            unpack=False,
+        )
+        return csv_path
+
+    def process_csv(self, csv_path: str, split: str) -> List[Instance]:
+        """Read and process a CSV file to generate instances."""
+        instances: List[Instance] = []
+        with open(csv_path, "r", encoding="utf-8") as f:
+            reader = csv.DictReader(f)
+            for row in reader:
+                # Validate required fields
+                if not row.get("question") or not row.get("answer_idx") or not row.get("opa"):
+                    print(f"Skipping invalid row: {row}")
+                    continue
+
+                # Map answers to indices
+                option_map = {
+                    "A": row.get("opa", "Not applicable"),
+                    "B": row.get("opb", "Not applicable"),
+                    "C": row.get("opc", "Not applicable"),
+                    "D": row.get("opd", "Not applicable"),
+                    "E": row.get("ope", "Not applicable"),
+                }
+
+                # Correct answer text
+                correct_option = row["answer_idx"]
+
+                # Build references using POSSIBLE_ANSWER_CHOICES
+                references = [
+                    Reference(
+                        Output(text=option_map.get(option, "Not applicable")),
+                        tags=[CORRECT_TAG] if option == correct_option else [],
+                    )
+                    for option in self.POSSIBLE_ANSWER_CHOICES
+                ]
+
+                # Create instance
+                instance = Instance(
+                    input=Input(text=row["question"]),
+                    references=references,
+                    split=split,
+                )
+                instances.append(instance)
+        return instances
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        """Download and process dataset to generate instances."""
+        instances: List[Instance] = []
+        for split_suffix, split in self.splits.items():
+            csv_path = self.download_csv(output_path, split_suffix)
+            instances.extend(self.process_csv(csv_path, split))
+        return instances

helm/benchmark/scenarios/medcalc_bench_scenario.py

@@ -0,0 +1,125 @@
+from typing import Dict, List
+from datasets import load_dataset
+
+from helm.common.hierarchical_logger import hlog
+from helm.benchmark.scenarios.scenario import (
+    Scenario,
+    Instance,
+    Reference,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    PassageQuestionInput,
+    Output,
+)
+
+
+class MedCalcBenchScenario(Scenario):
+    """
+    MedCalc-Bench is the first medical calculation dataset used to benchmark
+    LLMs ability to serve as clinical calculators.
+    Each instance in the dataset consists of a patient note, a question asking
+    to compute a specific clinical value, a final answer value, and a step-by-step
+    solution explaining how the final answer was obtained. Our dataset covers 55
+    different calculation tasks. We hope this dataset serves as a call to improve
+    the verbal and computational reasoning skills of LLMs in medical settings.
+
+    This dataset contains a training dataset of 10,053 instances and a testing
+    dataset of 1,047 instances.
+
+    Dataset: https://huggingface.co/datasets/ncbi/MedCalc-Bench-v1.0
+    Paper: https://arxiv.org/abs/2406.12036
+
+    Sample Prompt:
+        Given a patient note and a clinical question, compute the requested medical value.
+        Be as concise as possible.
+
+        Patient note: A 70-year-old female was rushed into the ICU due to respiratory distress,
+        following which she was promptly put on mechanical ventilation. Her delivered oxygen fell
+        to 51 % FiO₂; meanwhile, her partial pressure of oxygen (PaO₂) registered at 74 mm Hg.
+        She was conscious but visibly disoriented with a functional Glasgow Coma Score of 12.
+        She was hypotensive with blood pressure of 91/70 mm Hg. Multiple vasopressors are being administered
+        simultaneously including DOPamine at 4 mcg/kg/min, norEPINEPHrine at 0.06 mcg/kg/min,
+        DOBUTamine at 3 mcg/kg/min, and EPINEPHrine at 0.03 mcg/kg/min. Laboratory evaluations
+        revealed mild renal impairment with creatinine levels slightly elevated at 1.6 mg/dL
+        and a bilirubin level of 1.9 mg/dL. Her platelet count was found to be 165,000/µL.
+        Her daily urine output of 950 mL.
+        Question: What is the patient's Sequential Organ Failure Assessment (SOFA) Score?
+
+        Answer:
+
+    @misc{khandekar2024medcalcbench,
+        title={MedCalc-Bench: Evaluating Large Language Models for Medical Calculations},
+        author={
+            Nikhil Khandekar and Qiao Jin and Guangzhi Xiong and Soren Dunn and Serina S Applebaum and
+            Zain Anwar and Maame Sarfo-Gyamfi and Conrad W Safranek and Abid A Anwar and Andrew Zhang and
+            Aidan Gilson and Maxwell B Singer and Amisha Dave and Andrew Taylor and Aidong Zhang and
+            Qingyu Chen and Zhiyong Lu
+        },
+        year={2024},
+        eprint={2406.12036},
+        archivePrefix={arXiv},
+        primaryClass={
+            id='cs.CL' full_name='Computation and Language' is_active=True alt_name='cmp-lg'
+            in_archive='cs' is_general=False description='Covers natural language processing.
+            Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial
+            languages (programming languages, logics, formal systems) that does not explicitly
+            address natural-language issues broadly construed (natural-language processing, computational
+            linguistics, speech, text retrieval, etc.) is not appropriate for this area.'
+        }
+    }
+    """
+
+    name = "medcalc_bench"
+    description = (
+        "A dataset which consists of a patient note, a question "
+        "requesting to compute a specific medical value, and a ground truth answer."
+    )
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    def __init__(self):
+        super().__init__()
+
+    def process_csv(self, data, split: str) -> List[Instance]:
+        instances: List[Instance] = []
+        hlog(f"Processing data for {split} split")
+        for row in data:
+            question = row["Question"]
+            ground_truth_answer = row["Ground Truth Answer"]
+            patient_note = row["Patient Note"]
+            id = row["Row Number"]
+
+            prompt = PassageQuestionInput(
+                passage=patient_note + "\n", question=question + "\n", passage_prefix="Patient note: "
+            )
+
+            extra_data = {
+                "category": row["Category"],
+                "upper_limit": row["Upper Limit"],
+                "lower_limit": row["Lower Limit"],
+            }
+
+            instance = Instance(
+                input=prompt,
+                references=[Reference(Output(text=ground_truth_answer), tags=[CORRECT_TAG])],
+                extra_data=extra_data,
+                split=split,
+                id=id,
+            )
+            instances.append(instance)
+        return instances
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        # Load the MedCalc-Bench dataset from Hugging Face
+        dataset = load_dataset("ncbi/MedCalc-Bench-v1.0")
+
+        # Process all the instances - limit to zero shot setting
+        instances: List[Instance] = []
+        splits: Dict[str, str] = {
+            # "train": TRAIN_SPLIT,
+            "test": TEST_SPLIT,
+        }
+        for hf_split, split in splits.items():
+            data = dataset[hf_split]
+            instances.extend(self.process_csv(data, split))
+
+        return instances

helm/benchmark/scenarios/medec_scenario.py

@@ -0,0 +1,120 @@
+import csv
+import os
+from typing import List
+from helm.benchmark.scenarios.scenario import (
+    Scenario,
+    Instance,
+    Reference,
+    CORRECT_TAG,
+    TEST_SPLIT,
+    Input,
+    Output,
+)
+from helm.common.general import ensure_file_downloaded
+
+
+class MedecScenario(Scenario):
+    """
+    Processes the MEDEC dataset for medical error detection and correction tasks.
+
+    MEDEC is the first publicly available benchmark for medical error detection and correction in clinical notes,
+    introduced in "Ben Abacha et al., 2024." The dataset includes 3,848 clinical texts from the MS and UW collections,
+    covering five types of errors:
+    - Diagnosis
+    - Management
+    - Treatment
+    - Pharmacotherapy
+    - Causal Organism
+
+    The dataset consists of:
+    - Training Set: 2,189 MS texts
+    - Validation Set: 574 MS texts and 160 UW texts
+    - Test Set: 597 MS texts and 328 UW texts
+
+    Each clinical text is labeled as either correct or containing one error. The task involves:
+    (A) Predicting the error flag (1: the text contains an error, 0: the text has no errors).
+    (B) For flagged texts, extracting the sentence that contains the error.
+    (C) Generating a corrected sentence.
+
+    The MEDEC dataset was used for the MEDIQA-CORR shared task to evaluate seventeen participating systems.
+    Recent LLMs (e.g., GPT-4, Claude 3.5 Sonnet, Gemini 2.0 Flash) have been evaluated on this dataset, showing good
+    performance but still lagging behind medical doctors in error detection and correction tasks.
+
+    Task:
+    Given a clinical text, models must identify errors and correct them while demonstrating medical knowledge
+    and reasoning capabilities.
+    """
+
+    GIT_HASH = "49c59dcba43a7af8717590a405b11a57f42b04df"
+    TRAIN_URL = f"https://raw.githubusercontent.com/abachaa/MEDEC/{GIT_HASH}/MEDEC-MS/MEDEC-Full-TrainingSet-with-ErrorType.csv"  # noqa: E501
+    TEST_URL = f"https://raw.githubusercontent.com/abachaa/MEDEC/{GIT_HASH}/MEDEC-MS/MEDEC-MS-TestSet-with-GroundTruth-and-ErrorType.csv"  # noqa: E501
+
+    name = "medec"
+    description = "A dataset containing medical narratives with error detection and correction pairs."
+    tags = ["error_detection", "error_correction", "biomedical"]
+
+    def download_csv(self, url: str, output_path: str, file_name: str) -> str:
+        """Download the CSV file and return its path."""
+        csv_path = os.path.join(output_path, file_name)
+        ensure_file_downloaded(source_url=url, target_path=csv_path, unpack=False)
+        return csv_path
+
+    def process_csv(self, csv_path: str, split: str) -> List[Instance]:
+        """Read and process a CSV file to generate instances."""
+        instances: List[Instance] = []
+        with open(csv_path, "r", encoding="utf-8") as f:
+            reader = csv.DictReader(f)
+
+            for row in reader:
+                # Ensure required fields exist
+                if not row.get("Sentences"):
+                    continue
+
+                # Extract text content
+                input_text = row["Sentences"].strip()
+
+                # Extract metadata
+                error_flag = int(row.get("Error Flag", 0))
+                error_sentence_id = row.get("Error Sentence ID", "-1").strip()
+                corrected_sentence = row.get("Corrected Sentence", "").strip()
+
+                references = []
+                if error_flag == 1 and corrected_sentence != "NA" and error_sentence_id != "-1":
+                    # If there is an error, store the corrected sentence
+                    references.append(
+                        Reference(
+                            Output(text=f"{error_sentence_id} {corrected_sentence}"),
+                            tags=[CORRECT_TAG],
+                        )
+                    )
+                else:
+                    # If no error exists, expect "CORRECT" as the output
+                    references.append(
+                        Reference(
+                            Output(text="CORRECT"),
+                            tags=[CORRECT_TAG],
+                        )
+                    )
+
+                instance = Instance(
+                    input=Input(text=input_text),
+                    references=references,
+                    split=split,
+                )
+                instances.append(instance)
+
+        return instances
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        """Download and process the dataset to generate instances."""
+        instances: List[Instance] = []
+
+        # Download and process the training set - limit to zero shot setting for now
+        # train_csv = self.download_csv(self.TRAIN_URL, output_path, "medec_train.csv")
+        # instances.extend(self.process_csv(train_csv, TRAIN_SPLIT))
+
+        # Download and process the test set
+        test_csv = self.download_csv(self.TEST_URL, output_path, "medec_test.csv")
+        instances.extend(self.process_csv(test_csv, TEST_SPLIT))
+
+        return instances

helm/benchmark/scenarios/medhallu_scenario.py

@@ -0,0 +1,66 @@
+from typing import List
+from datasets import load_dataset
+
+from helm.benchmark.scenarios.scenario import (
+    Scenario,
+    Instance,
+    Reference,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Output,
+    Input,
+)
+
+
+class MedHalluScenario(Scenario):
+    """
+    MedHallu is a medical hallucination dataset that consists of PubMed articles and associated questions,
+    with the objective being to classify whether the answer is factual or hallucinated.
+    MedHallu: https://medhallu.github.io/
+    """
+
+    name = "medhallu"
+    description = "A dataset of PubMed articles and associated questions, with the objective being to classify whether the answer is factual or hallucinated."  # noqa: E501
+    tags = ["knowledge", "reasoning", "biomedical"]
+
+    def create_instance(self, question, knowledge, answer, label, split):
+        prompt_text = f"""World Knowledge: {knowledge}
+
+Question: {question}
+
+Answer: {answer}
+"""
+        return Instance(
+            input=Input(
+                text=prompt_text,
+            ),
+            references=[Reference(Output(text=label), tags=[CORRECT_TAG])],
+            split=split,
+        )
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        # Load the MedCalc-Bench dataset from Hugging Face
+        dataset = load_dataset(
+            "UTAustin-AIHealth/MedHallu",
+            "pqa_labeled",
+            split="train",
+            revision="515060458a945c633debc6fd5baac7764416b724",
+        )
+
+        # Process all the instances - limit to zero shot setting
+        instances: List[Instance] = []
+        for row in dataset:
+            # print("printing row", row)
+            question = row["Question"]
+            ground_truth_answer = row["Ground Truth"]
+            knowledge = row["Knowledge"]
+            hallucinated_answer = row["Hallucinated Answer"]
+
+            gt_instance = self.create_instance(question, knowledge, ground_truth_answer, label="0", split=TEST_SPLIT)
+            instances.append(gt_instance)
+
+            hallucinated_instance = self.create_instance(
+                question, knowledge, hallucinated_answer, label="1", split=TEST_SPLIT
+            )
+            instances.append(hallucinated_instance)
+        return instances

helm/benchmark/scenarios/medi_qa_scenario.py

@@ -0,0 +1,105 @@
+from typing import Dict, List
+from datasets import load_dataset
+
+from helm.common.hierarchical_logger import hlog
+from helm.benchmark.scenarios.scenario import (
+    Scenario,
+    Instance,
+    Reference,
+    TEST_SPLIT,
+    CORRECT_TAG,
+    Input,
+    Output,
+)
+
+
+class MediQAScenario(Scenario):
+    """
+    MEDIQA-QA is a dataset designed to benchmark large language models (LLMs) on medical
+    question answering (QA) tasks.
+    Each instance in the dataset includes a medical question, a set of candidate answers,
+    relevance annotations for ranking, and additional context to evaluate understanding
+    and retrieval capabilities in a healthcare setting.
+
+    The dataset encompasses diverse question types, including consumer health queries
+    and clinical questions, making it suitable for assessing LLMs' ability to answer
+    consumer healthcare questions.
+
+    This dataset comprises two training sets of 104 instances each, a validation set
+    of 25 instances, and a testing set of 150 instances.
+
+    Dataset: https://huggingface.co/datasets/bigbio/mediqa_qa
+    Paper: https://aclanthology.org/W19-5039/
+
+    Sample Prompt:
+        Answer the following consumer health question.
+
+        Question: Noonan syndrome. What are the references with noonan syndrome
+        and polycystic renal disease?
+        Answer:
+
+    @inproceedings{MEDIQA2019,
+        author    = {Asma {Ben Abacha} and Chaitanya Shivade and Dina Demner{-}Fushman},
+        title     = {Overview of the MEDIQA 2019 Shared Task on Textual Inference,
+                     Question Entailment and Question Answering},
+        booktitle = {ACL-BioNLP 2019},
+        year      = {2019}
+    }
+    """
+
+    name = "medi_qa"
+    description = (
+        "A dataset including a medical question, a set of candidate answers,"
+        "relevance annotations for ranking, and additional context to evaluate understanding"
+        "and retrieval capabilities in a healthcare setting."
+    )
+    tags = ["knowledge", "biomedical"]
+
+    def _get_highest_ranked_answer(self, answers: List[Dict[str, Dict[str, str]]]) -> str:
+        best_answer: str = ""
+        for answer in answers:
+            if answer["Answer"]["ReferenceRank"] != 1:
+                continue
+            best_answer = answer["Answer"]["AnswerText"]
+            break
+        return best_answer
+
+    def process_csv(self, data, split: str) -> List[Instance]:
+        instances: List[Instance] = []
+        hlog(f"Processing data for {split} split")
+        total_tokens: int = 0
+        counter = 0
+        for row in data:
+            row = row["QUESTION"]
+            question = row["QuestionText"]
+            ground_truth_answer = self._get_highest_ranked_answer(row["AnswerList"])
+            id = row["QID"]
+            counter += 1
+            total_tokens += len(ground_truth_answer.split())
+            instances.append(
+                Instance(
+                    input=Input(question),
+                    references=[Reference(Output(ground_truth_answer), tags=[CORRECT_TAG])],
+                    split=split,
+                    id=id,
+                )
+            )
+        return instances
+
+    def get_instances(self, output_path: str) -> List[Instance]:
+        # Load the MEDIQA dataset from Hugging Face
+        dataset = load_dataset("bigbio/mediqa_qa")
+
+        # Process all the instances
+        instances: List[Instance] = []
+        # Limit to zero shot setting
+        splits: Dict[str, str] = {
+            # "train_live_qa_med": TRAIN_SPLIT,
+            # "validation": VALID_SPLIT,
+            "test": TEST_SPLIT,
+        }
+        for hf_split, split in splits.items():
+            data = dataset[hf_split]
+            instances.extend(self.process_csv(data, split))
+
+        return instances

helm/benchmark/scenarios/medication_qa_scenario.py

@@ -5,7 +5,7 @@ import pandas as pd
 
 from helm.common.general import ensure_file_downloaded
 
-from .scenario import CORRECT_TAG, TEST_SPLIT, Input, Instance, Output, Reference, Scenario
+from helm.benchmark.scenarios.scenario import CORRECT_TAG, TEST_SPLIT, Input, Instance, Output, Reference, Scenario
 
 
 class MedicationQAScenario(Scenario):
@@ -31,7 +31,7 @@ class MedicationQAScenario(Scenario):
     FILENAME = "MedInfo2019-QA-Medications.xlsx"
 
     name = "medication_qa"
-    description = "MedInfo 2019 MedicationQA medication question answering task"
+    description = "Open text question-answer pairs regarding consumer health questions about medication."
     tags = ["knowledge", "generation", "question_answering", "biomedical"]
 
     def download_medication_qa(self, path: str):