Run pyfmt on .github and tools (#6165)

run `lintrunner -a --all-files --take PYFMT`

Update pyproject.toml with a few options for usort that make it more
similar to pytorch

I can break this down into smaller PRs if this is too large

Author: clee2000

.github/scripts/benchmarks/gather_metadata.py

@@ -5,8 +5,8 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-import os
 import json
+import os
 import time
 from typing import Any
 

.github/scripts/get_tutorials_stats.py

@@ -10,6 +10,7 @@
 
 import boto3  # type: ignore[import]
 
+
 METADATA_PATH = "ossci_tutorials_stats/metadata.csv"
 FILENAMES_PATH = "ossci_tutorials_stats/filenames.csv"
 

.github/scripts/update_commit_hashes.py

@@ -6,6 +6,7 @@
 
 import requests
 
+
 UPDATEBOT_TOKEN = os.environ["UPDATEBOT_TOKEN"]
 PYTORCHBOT_TOKEN = os.environ["PYTORCHBOT_TOKEN"]
 

.github/scripts/upload_benchmark_results.py

@@ -14,13 +14,13 @@
 from argparse import Action, ArgumentParser, Namespace
 from decimal import Decimal
 from json.decoder import JSONDecodeError
-
 from logging import info
 from typing import Any, Callable, Dict, List, Optional
 from warnings import warn
 
 import boto3
 
+
 logging.basicConfig(level=logging.INFO)
 
 

.github/scripts/validate_scale_config.py

@@ -9,16 +9,14 @@
 import copy
 import json
 import os
-
 import urllib.request
 from pathlib import Path
-
 from typing import Any, cast, Dict, List, NamedTuple
 
 import jsonschema
-
 import yaml
 
+
 MAX_AVAILABLE_MINIMUM = 50
 
 # Paths relative to their respective repositories

.lintrunner.toml

@@ -266,10 +266,8 @@ include_patterns = [
     '**/*.py',
 ]
 exclude_patterns = [
-    '.github/scripts/**',
     'aws/lambda/**',
     's3_management/**',
-    'tools/**',
 ]
 command = [
     'python3',

pyproject.toml

@@ -14,6 +14,12 @@ requires = [
 # Use legacy backend to import local packages in setup.py
 build-backend = "setuptools.build_meta:__legacy__"
 
+[tool.isort]
+lines_after_imports = 2
+multi_line_output = 3
+indent = 4
+include_trailing_comma = true
+combine_as_imports = true
 
 [tool.black]
 # Uncomment if pyproject.toml worked fine to ensure consistency with flake8

tools/analytics/cubinsizes.py

@@ -12,27 +12,30 @@
 try:
     from elftools.elf.elffile import ELFFile
 except ModuleNotFoundError:
-    print(f'elftools module not found, trying to install it from pip')
+    print(f"elftools module not found, trying to install it from pip")
     from pip._internal import main as pip_main
+
     try:
         pip_main(["install", "pyelftools", "--user"])
     except SystemExit:
-        print(f'PIP installation failed, please install it manually by invoking "{sys.executable} -mpip install pyelftools --user"')
+        print(
+            f'PIP installation failed, please install it manually by invoking "{sys.executable} -mpip install pyelftools --user"'
+        )
         sys.exit(-1)
     from elftools.elf.elffile import ELFFile
 
 
 # From https://stackoverflow.com/questions/1094841/reusable-library-to-get-human-readable-version-of-file-size
-def sizeof_fmt(num, suffix='B'):
-    for unit in ['', 'Ki', 'Mi', 'Gi', 'Ti', 'Pi', 'Ei', 'Zi']:
+def sizeof_fmt(num, suffix="B"):
+    for unit in ["", "Ki", "Mi", "Gi", "Ti", "Pi", "Ei", "Zi"]:
         if abs(num) < 1024.0:
             return "%3.1f%s%s" % (num, unit, suffix)
         num /= 1024.0
-    return "%.1f%s%s" % (num, 'Yi', suffix)
+    return "%.1f%s%s" % (num, "Yi", suffix)
 
 
-def compute_cubin_sizes(file_name, section_name='.nv_fatbin', debug=False):
-    with open(file_name, 'rb') as f:
+def compute_cubin_sizes(file_name, section_name=".nv_fatbin", debug=False):
+    with open(file_name, "rb") as f:
         elf_file = ELFFile(f)
         nv_fatbin = elf_file.get_section_by_name(section_name)
         if nv_fatbin is None:
@@ -41,20 +44,32 @@ def compute_cubin_sizes(file_name, section_name='.nv_fatbin', debug=False):
         idx, offs = 0, 0
         elf_sizes = {}
         while offs < len(data):
-            (magic, version, header_size, fatbin_size) = struct.unpack('IHHL', data[offs: offs + 16])
-            if magic != 0xba55ed50 or version != 1:
-                raise RuntimeError(f"Unexpected fatbin magic {hex(magic)} or version {version}")
+            (magic, version, header_size, fatbin_size) = struct.unpack(
+                "IHHL", data[offs : offs + 16]
+            )
+            if magic != 0xBA55ED50 or version != 1:
+                raise RuntimeError(
+                    f"Unexpected fatbin magic {hex(magic)} or version {version}"
+                )
             if debug:
-                print(f"Found fatbin at {offs}  header_size={header_size} fatbin_size={fatbin_size}")
+                print(
+                    f"Found fatbin at {offs}  header_size={header_size} fatbin_size={fatbin_size}"
+                )
             offs += header_size
             fatbin_end = offs + fatbin_size
             while offs < fatbin_end:
-                (kind, version, hdr_size, elf_size, empty, code_ver, sm_ver) = struct.unpack('HHILLIH', data[offs: offs + 30])
+                (kind, version, hdr_size, elf_size, empty, code_ver, sm_ver) = (
+                    struct.unpack("HHILLIH", data[offs : offs + 30])
+                )
                 if version != 0x0101 or kind not in [1, 2]:
-                    raise RuntimeError(f"Unexpected cubin version {hex(version)} or kind {kind}")
+                    raise RuntimeError(
+                        f"Unexpected cubin version {hex(version)} or kind {kind}"
+                    )
                 sm_ver = f'{"ptx" if kind == 1 else "sm"}_{sm_ver}'
                 if debug:
-                    print(f"    {idx}: elf_size={elf_size} code_ver={hex(code_ver)} sm={sm_ver}")
+                    print(
+                        f"    {idx}: elf_size={elf_size} code_ver={hex(code_ver)} sm={sm_ver}"
+                    )
                 if sm_ver not in elf_sizes:
                     elf_sizes[sm_ver] = 0
                 elf_sizes[sm_ver] += elf_size
@@ -71,7 +86,7 @@ def __init__(self, ar_name: str) -> None:
     def __enter__(self) -> str:
         self._pwd = os.getcwd()
         rc = self._tmpdir.__enter__()
-        subprocess.check_call(['ar', 'x', self.ar_name])
+        subprocess.check_call(["ar", "x", self.ar_name])
         return rc
 
     def __exit__(self, ex, value, tb) -> None:
@@ -86,40 +101,44 @@ def dict_add(rc: Dict[str, int], b: Dict[str, int]) -> Dict[str, int]:
 
 
 def main():
-    if sys.platform != 'linux':
-        print('This script only works with Linux ELF files')
+    if sys.platform != "linux":
+        print("This script only works with Linux ELF files")
         return
     if len(sys.argv) < 2:
-        print(f"{sys.argv[0]} invoked without any arguments trying to infer location of libtorch_cuda")
+        print(
+            f"{sys.argv[0]} invoked without any arguments trying to infer location of libtorch_cuda"
+        )
         import torch
-        fname = os.path.join(os.path.dirname(torch.__file__), 'lib', 'libtorch_cuda.so')
+
+        fname = os.path.join(os.path.dirname(torch.__file__), "lib", "libtorch_cuda.so")
     else:
         fname = sys.argv[1]
 
     if not os.path.exists(fname):
         print(f"Can't find {fname}")
         sys.exit(-1)
 
-    section_names = ['.nv_fatbin', '__nv_relfatbin']
+    section_names = [".nv_fatbin", "__nv_relfatbin"]
     results = {name: {} for name in section_names}
     print(f"Analyzing {fname}")
-    if os.path.splitext(fname)[1] == '.a':
+    if os.path.splitext(fname)[1] == ".a":
         with ArFileCtx(fname):
             for fname in os.listdir("."):
-                if not fname.endswith(".o"): continue
+                if not fname.endswith(".o"):
+                    continue
                 for section_name in section_names:
                     elf_sizes = compute_cubin_sizes(fname, section_name)
                     dict_add(results[section_name], elf_sizes)
     else:
-        for section_name in ['.nv_fatbin', '__nv_relfatbin']:
+        for section_name in [".nv_fatbin", "__nv_relfatbin"]:
             dict_add(results[section_name], compute_cubin_sizes(fname, section_name))
 
     for section_name in section_names:
         elf_sizes = results[section_name]
         print(f"{section_name} size {sizeof_fmt(sum(elf_sizes.values()))}")
-        for (sm_ver, total_size) in elf_sizes.items():
+        for sm_ver, total_size in elf_sizes.items():
             print(f"  {sm_ver}: {sizeof_fmt(total_size)}")
 
 
-if __name__ == '__main__':
+if __name__ == "__main__":
     main()

tools/analytics/download_count_wheels.py

@@ -1,16 +1,18 @@
-from collections import defaultdict
-from datetime import datetime, timedelta, timezone
 import gzip
 import os
 import re
 import urllib
+from collections import defaultdict
+from datetime import datetime, timedelta, timezone
 
-from tqdm import tqdm
 import boto3
+from tqdm import tqdm
+
+
+S3 = boto3.resource("s3")
+CLIENT = boto3.client("s3")
+BUCKET = S3.Bucket("pytorch")
 
-S3 = boto3.resource('s3')
-CLIENT = boto3.client('s3')
-BUCKET = S3.Bucket('pytorch')
 
 class CacheEntry:
     _size = None
@@ -38,66 +40,56 @@ def target_arch(self) -> str:
 
     @property
     def package_name(self) -> str:
-        filename_contents = os.path.basename(self.download_uri).split('-')
+        filename_contents = os.path.basename(self.download_uri).split("-")
         return filename_contents[0]
 
     @property
     def package_version(self) -> str:
         if "dev" in self.download_uri:
-            results = re.search(
-                r"[0-9]+\.[0-9]+\.[0-9]+\.dev[0-9]+",
-                self.download_uri
-            )
+            results = re.search(r"[0-9]+\.[0-9]+\.[0-9]+\.dev[0-9]+", self.download_uri)
         else:
-            results = re.search(
-                r"[0-9]+\.[0-9]+\.[0-9]+", self.download_uri
-            )
+            results = re.search(r"[0-9]+\.[0-9]+\.[0-9]+", self.download_uri)
         if not results:
             raise Exception("Wtf there's no version o.O")
         return results[0]
 
     @property
     def size(self) -> int:
         if self._size is None:
-            for key in BUCKET.objects.filter(
-                    Prefix=self.download_uri.lstrip("/")
-            ):
+            for key in BUCKET.objects.filter(Prefix=self.download_uri.lstrip("/")):
                 self._size = key.size
             if self._size is None:
-                raise Exception(
-                    f"No object found for prefix {self.download_uri}"
-                )
+                raise Exception(f"No object found for prefix {self.download_uri}")
         return self._size
 
     @property
     def downloads(self):
         return self.bytes_sent // self.size
 
+
 def parse_logs(log_directory: str) -> dict:
     bytes_cache = {}
-    for (dirpath, _, filenames) in os.walk(log_directory):
+    for dirpath, _, filenames in os.walk(log_directory):
         for filename in tqdm(filenames):
-            with gzip.open(os.path.join(dirpath, filename), 'r') as gf:
+            with gzip.open(os.path.join(dirpath, filename), "r") as gf:
                 string = gf.read().decode("utf-8")
                 entries = []
                 entries += string.splitlines()[2:]
             for entry in entries:
-                columns = entry.split('\t')
+                columns = entry.split("\t")
                 bytes_sent = int(columns[3])
-                download_uri = urllib.parse.unquote(
-                    urllib.parse.unquote(columns[7])
-                )
+                download_uri = urllib.parse.unquote(urllib.parse.unquote(columns[7]))
                 status = columns[8]
-                if not all([
-                        status.startswith("2"),
-                        download_uri.endswith((".whl", ".zip"))
-                ]):
+                if not all(
+                    [status.startswith("2"), download_uri.endswith((".whl", ".zip"))]
+                ):
                     continue
                 if not bytes_cache.get(download_uri):
                     bytes_cache[download_uri] = CacheEntry(download_uri)
                 bytes_cache[download_uri].bytes_sent += bytes_sent
     return bytes_cache
 
+
 def output_results(bytes_cache: dict) -> None:
     os_results = defaultdict(int)
     arch_results = defaultdict(int)
@@ -106,25 +98,19 @@ def output_results(bytes_cache: dict) -> None:
         try:
             os_results[val.os_type] += val.downloads
             arch_results[val.target_arch] += val.downloads
-            package_results[val.package_name][val.package_version] += (
-                val.downloads
-            )
+            package_results[val.package_name][val.package_version] += val.downloads
         except Exception:
             pass
     print("=-=-= Results =-=-=")
     print("=-=-= OS =-=-=")
     total_os_num = sum(os_results.values())
     for os_type, num in os_results.items():
-        print(
-            f"\t* {os_type}: {num} ({(num/total_os_num)*100:.2f}%)"
-        )
+        print(f"\t* {os_type}: {num} ({(num/total_os_num)*100:.2f}%)")
 
     print("=-=-= ARCH =-=-=")
     total_arch_num = sum(arch_results.values())
     for arch_type, num in arch_results.items():
-        print(
-            f"\t* {arch_type}: {num} ({(num/total_arch_num) * 100:.2f}%)"
-        )
+        print(f"\t* {arch_type}: {num} ({(num/total_arch_num) * 100:.2f}%)")
 
     print("=-=-= By Package =-=-=")
     for package_name, upper_val in package_results.items():
@@ -135,11 +121,14 @@ def output_results(bytes_cache: dict) -> None:
                 f"\t* {package_version}: {num} ({(num/total_package_num) * 100:.2f}%)"
             )
 
+
 def download_logs(log_directory: str, since: float):
     dt_now = datetime.now(timezone.utc)
     dt_end = datetime(dt_now.year, dt_now.month, dt_now.day, tzinfo=timezone.utc)
-    dt_start = dt_end - timedelta(days=1, hours=1) # Add 1 hour padding to account for potentially missed logs due to timing
-    for key in tqdm(BUCKET.objects.filter(Prefix='cflogs')):
+    dt_start = dt_end - timedelta(
+        days=1, hours=1
+    )  # Add 1 hour padding to account for potentially missed logs due to timing
+    for key in tqdm(BUCKET.objects.filter(Prefix="cflogs")):
         remote_fname = key.key
         local_fname = os.path.join(log_directory, remote_fname)
         # Only download things from yesterday
@@ -156,8 +145,8 @@ def download_logs(log_directory: str, since: float):
 
 if __name__ == "__main__":
     print("Downloading logs")
-    download_logs('cache', 1)
+    download_logs("cache", 1)
     print("Parsing logs")
-    cache = parse_logs('cache/cflogs/')
+    cache = parse_logs("cache/cflogs/")
     print("Calculating results")
     output_results(cache)