MMLU-by-task-Leaderboard / details_data_processor.py
Corey Morris
shortened file name
38d88f9
raw
history blame
4.81 kB
import pandas as pd
import os
import fnmatch
import json
import re
import numpy as np
import requests
from urllib.parse import quote
from datetime import datetime
import uuid
class DetailsDataProcessor:
# Download
#url example https://huggingface.co/datasets/open-llm-leaderboard/details/resolve/main/64bits/LexPodLM-13B/details_harness%7ChendrycksTest-moral_scenarios%7C5_2023-07-25T13%3A41%3A51.227672.json
def __init__(self, directory='results', pattern='results*.json'):
self.directory = directory
self.pattern = pattern
def _find_files(self, directory='results', pattern='results*.json'):
matching_files = [] # List to hold matching filenames
for root, dirs, files in os.walk(directory):
for basename in files:
if fnmatch.fnmatch(basename, pattern):
filename = os.path.join(root, basename)
matching_files.append(filename) # Append the matching filename to the list
return matching_files # Return the list of matching filenames
# @staticmethod
# def download_file(url, directory='details_data'):
# # Define the prefix to be removed from the URL
# url_prefix = "https://huggingface.co/datasets/open-llm-leaderboard/details/resolve/main/"
# # Remove the prefix from the URL
# file_name_part = url.replace(url_prefix, '')
# # Replace characters that don't play nice with file systems
# safe_file_name = re.sub(r'[<>:"/\\|?*]', '_', file_name_part) # Replace with '_'
# save_file_path = os.path.join(directory, safe_file_name)
# error_count = 0
# success_count = 0
# try:
# # Sending a GET request
# r = requests.get(url, allow_redirects=True)
# r.raise_for_status()
# # Writing the content to the specified file
# with open(save_file_path, 'wb') as file:
# file.write(r.content)
# success_count += 1
# except requests.ConnectionError as e:
# error_count += 1
# except requests.HTTPError as e:
# error_count += 1
# except FileNotFoundError as e:
# error_count += 1
# except Exception as e:
# error_count += 1
# return error_count, success_count
@staticmethod
def download_file(url, directory='details_data'):
# Extract relevant parts from the URL
segments = url.split('/')
organization = segments[-3]
model_name = segments[-2]
task = segments[-1].split('_')[0] # Assuming task is part of the last segment
# Construct the filename
safe_file_name = f"{organization}_{model_name}_{task}.json"
# Create the full save file path
save_file_path = os.path.join(directory, safe_file_name)
error_count = 0
success_count = 0
try:
# Sending a GET request
r = requests.get(url, allow_redirects=True)
r.raise_for_status()
# Writing the content to the specified file
with open(save_file_path, 'wb') as file:
file.write(r.content)
print(save_file_path)
success_count += 1
except requests.ConnectionError as e:
error_count += 1
except requests.HTTPError as e:
error_count += 1
except FileNotFoundError as e:
error_count += 1
except Exception as e:
error_count += 1
return error_count, success_count
@staticmethod
def single_file_pipeline(url, filename):
DetailsDataProcessor.download_file(url, filename)
# read file
with open(filename) as f:
data = json.load(f)
# convert to dataframe
df = pd.DataFrame(data)
return df
@staticmethod
def build_url(file_path):
segments = file_path.split('/')
bits = segments[1]
model_name = segments[2]
try:
timestamp = segments[3].split('_')[1]
except IndexError:
print(f"Error: {file_path}")
return None
url = f'https://huggingface.co/datasets/open-llm-leaderboard/details/resolve/main/{bits}/{model_name}/details_harness%7ChendrycksTest-moral_scenarios%7C5_{quote(timestamp, safe="")}'
return url
def pipeline(self):
dataframes = []
file_paths = self._find_files(self.directory, self.pattern)
for file_path in file_paths:
print(file_path)
url = self.generate_url(file_path)
file_path = file_path.split('/')[-1]
df = self.single_file_pipeline(url, file_path)
dataframes.append(df)
return dataframes