HUGS_energy / code /make_pretty_dataset.py
meg's picture
meg HF staff
Create code/make_pretty_dataset.py
5a65846 verified
import os
import json
from huggingface_hub import HfApi
import glob
from datetime import datetime
from datasets import Dataset
TOKEN = os.environ.get("HF_WRITE_TOKEN")
API = HfApi(token=TOKEN)
REPO_ID = "AIEnergyScore/results_debug" #"meg/calculate_carbon_runs"
UPLOAD_REPO_ID = 'meg/HUGS_energy'
output_directory = API.snapshot_download(repo_id=REPO_ID, repo_type='dataset')
print(output_directory)
#runs_dir = glob.glob(f"{output_directory}/*")
#print(runs_dir)
dataset_results = []
for task in ['text_generation']:
org_dirs = glob.glob(f"{output_directory}/{task}/*") #runs/{task}/*")
print(org_dirs)
for org_dir in org_dirs:
org = org_dir.split("/")[-1]
model_dirs = glob.glob(f"{org_dir}/*")
print(model_dirs)
for model_dir in model_dirs:
model = model_dir.split("/")[-1]
model_runs = glob.glob(f"{model_dir}/*")
dates = [dir.split("/")[-1] for dir in model_runs]
try:
# Sort dates as dates
sorted_dates = sorted(
[datetime.strptime(date, '%Y-%m-%d-%H-%M-%S') for date in
dates])
# Convert back to string format
sorted_dates_str = [date.strftime('%Y-%m-%d-%H-%M-%S') for date in
sorted_dates]
last_date = sorted_dates_str[-1]
most_recent_run = f"{model_dir}/{last_date}"
print(most_recent_run)
try:
benchmark_report = json.loads(open(f"{most_recent_run}/benchmark_report.json", "rb+").read())
print(benchmark_report)
prefill_data = benchmark_report['prefill']
prefill_energy = prefill_data['energy']
prefill_efficiency = prefill_data['efficiency']
decode_data = benchmark_report['decode']
decode_energy = decode_data['energy']
decode_efficiency = decode_data['efficiency']
preprocess_data = benchmark_report['preprocess']
preprocess_energy = preprocess_data['energy']
preprocess_efficiency = preprocess_data['efficiency']
dataset_results += [{'task':task, 'org':org, 'model':model, 'hardware':'a10g-large',
'date':last_date, 'prefill':{'energy':prefill_energy,
'efficency':prefill_efficiency},
'decode':{'energy':decode_energy, 'efficiency':decode_efficiency},
'preprocess': {'energy':preprocess_energy, 'efficiency': preprocess_efficiency}},]
except FileNotFoundError:
error_report = open(f"{most_recent_run}/error.log", "rb+").read()
print(error_report)
except ValueError:
# Not a directory with a timestamp.
continue
hub_dataset_results = Dataset.from_list(dataset_results)
print(hub_dataset_results)
hub_dataset_results.push_to_hub(UPLOAD_REPO_ID, token=TOKEN)