# Dataset Format ## `med-student.json` Each entry in the JSON file is structured as follows: - **`unique_id`**: A unique identifier to distinguish each entry. - `section`: The type of content, which can be one of the following: - `qa`: Question-answer section. - `physical_exam`: Physical examination section. - `closure`: Closure section. - `diagnosis`: Diagnosis section. - `case_id`: The identifier of the case number, ranging from 1 to 44. - `conversation_turn_id`: Indicates the sequence of dialogue for the `qa` section. All other sections have `conversation_turn_id` with 1. - `input`: An object containing the input variable name and input data. - `ground_truth_output`: The expected response or output from the examiner. - `prompt`: An object containing the prompt: - `_type`: A string indicate this a prompt - `input_variables`: An array of input variable names used in the prompt template. - `template`: A prompt template that includes placeholders for the input variables. - `output`: An object containing the running model name and output data generate from that model. ### Example Case Object ``` { "unique_id": 1, "section": "qa", "case_id": 1, "conversation_turn_id": 1, "input": { "opening": "Opening Scenario:\n\...", "chat_history": "N/A" }, "ground_truth_output": "\"What con...", "prompt": { "_type": "prompt", "input_variables": [ "opening", "chat_history" ], "template": "You are a doctor an..." }, "output": { "gpt-3.5-turbo-1106": "When did ...", "gpt-4-1106-preview": "Can you d...", "claude-3-sonnet-20240229": "Can...", "claude-3-haiku-20240307": "Can ...", "claude-3-opus-20240229": "What ..." } } ``` ## `med-exam.json` Each entry in the JSON file is structured as follows: - **`unique_id`**: A unique identifier to distinguish each entry. - `section`: The type of content, which can be one of the following: - `qa`: Question-answer section. - `physical_exam`: Physical examination section. - `closure`: Closure section. - `diagnosis`: Diagnosis section. - `case_id`: The identifier of the case number, ranging from 1 to 44. - `conversation_turn_id`: Indicates the sequence of dialogue for the `qa` section. All other sections have `conversation_turn_id` with 1. - `input`: An object containing the input data for evaluation, also include the generated output data from student task. - `prompt`: An object containing the prompt: - `_type`: A string indicate this a prompt - `input_variables`: An array of input variable names used in the prompt template. - `template`: A prompt template that includes placeholders for the input variables. - `result`: An object containing the model name for the input data, examiner model name, and the evaluation result. ### Example Case Object ``` { "unique_id": 1, "section": "qa", "case_id": 1, "conversation_turn_id": 1, "input": { "prev_conversation": "N/A", "opening": "Opening Scenario:\n\nJosep...", "question": { "gpt-4o": "Can you describe what the...", "gpt-4-1106-preview": "Can you descr...", "gpt-3.5-turbo-1106": "When did the ...", "claude-3-opus-20240229": "What brin...", "claude-3-sonnet-20240229": "Can you...", "claude-3-haiku-20240307": "Can you ..." }, "ground_truth": "\"What concerns you m..." }, "prompt": { "_type": "prompt", "input_variables": [ "prev_conversation", "opening", "question", "ground_truth" ], "template": "As an evaluator for the U..." }, "output": { "gpt-4o": { "gpt-4-1106-preview": "Score: 1\nRea...", }, "gpt-4-1106-preview": { "gpt-4-1106-preview": "Score: 1\nRea...", }, "gpt-3.5-turbo-1106": { "gpt-4-1106-preview": "Score: 1\nRea...", }, "claude-3-opus-20240229": { "gpt-4-1106-preview": "Score: 0\nRea...", }, "claude-3-sonnet-20240229": { "gpt-4-1106-preview": "Score: 1\nRea...", }, "claude-3-haiku-20240307": { "gpt-4-1106-preview": "Score: 1\nRea..." } } }, ```