Generate results for llama3 benchmarks on azure

import re
import pandas as pd
import os
results_dir = r'C:\code\azureml-assets-sagoswami-llama3_1_assets\assets\evaluation_results'
dirs = ['gpt-4o', 'meta-llama-3', 'meta-llama3']
folders = os.listdir(results_dir)
out_folders = []
for folder in folders:
    for d in dirs:
        if d in folder:
            out_folders.append(folder)
            continue

def parse_evaluation_results():

    sections = []
    for folder in out_folders:
        file = os.path.join(results_dir, folder, 'spec.yaml')
        with open(file, 'r') as file:
            text = file.read()
        sections.append(text)
    results = []

    for section in sections:
        model_name_match = re.search(r"model_name:\s*(\S+)", section)
        dataset_name_match = re.search(r"dataset_name:\s*(\S+)", section)
        accuracy_match = re.search(r"accuracy:\s*([\d.]+)", section)

        if model_name_match and dataset_name_match and accuracy_match:
            result = {
                "model_name": model_name_match.group(1),
                "dataset_name": dataset_name_match.group(1),
                "accuracy": float(accuracy_match.group(1))
            }
            results.append(result)

    return results


def convert_to_markdown_table(results):
    # Create a dictionary to hold the data
    data = {}

    for result in results:
        model_name = result["model_name"]
        dataset_name = result["dataset_name"]
        accuracy = f"{result['accuracy']:.3f}"

        if dataset_name not in data:
            data[dataset_name] = {}
        data[dataset_name][model_name] = accuracy

    # Convert the dictionary to a DataFrame
    df = pd.DataFrame(data).T

    # Convert the DataFrame to a Markdown table
    markdown_table = df.to_markdown()

    return markdown_table


# Example usage
file_path = r"C:\code\results.txt"
results = parse_evaluation_results(file_path)
markdown_table = convert_to_markdown_table(results)
print(markdown_table)