chore(sec): added a handful of security checks

examples/pytorch/transformers_serve_cb_eval_job.py

@@ -186,7 +186,7 @@ def main():
     print("=" * 70)
 
     # Start server with output going directly to stdout/stderr
-    server_process = subprocess.Popen(serve_cmd, stdout=None, stderr=None)
+    server_process = subprocess.Popen(serve_cmd, stdout=None, stderr=None)  # noqa: S603
 
     wait_for_server_up(server_process, port=args.port, timeout=600)
 

pyproject.toml

@@ -26,7 +26,19 @@ select = [
     "UP",  # pyupgrade
     "FURB",  # refurb
     "SIM",  # flake8-simplify
-    "S110",  # bandit's try-except-pass rule
+    "S102",  # Use of `exec` detected
+    "S103",  # `os.chmod` setting a permissive mask on file or directory
+    "S104",  # Possible binding to all interfaces
+    "S108",  # Probable insecure usage of temporary file or directory
+    "S306",  # Use of `tempfile.mktemp` (use `tempfile.mkstemp` instead)
+    "S307",  # Use of `eval` detected
+    "S603",  # `subprocess` call: check for execution of untrusted input
+    "S604",  # Function call with `shell=True` parameter identified
+    "S605",  # Starting a process with a shell, possible injection
+    "S606",  # Starting a process without a shell
+    "S607",  # Starting a process with a partial executable path
+    "S110",  # `try`-`except`-`pass` detected, consider logging the exception
+    "S112",  # `try`-`except`-`continue` detected, consider logging the exception
     "C4",  # flake8-comprehensions
     "RUF013",  # Checks for the use of implicit Optional in type annotations when the default parameter value is None.
     "PERF102",  # Checks for uses of dict.items() that discard either the key or the value when iterating over the dictionary.
@@ -51,6 +63,9 @@ extend-safe-fixes = [
 "__init__.py" = ["E402", "F401", "F403", "F811"]
 "src/transformers/file_utils.py" = ["F401"]
 "src/transformers/utils/dummy_*.py" = ["F401"]
+"tests/**/*.py" = ["S104", "S603", "S607"]
+"benchmark/**/*.py" = ["S603", "S607"]
+"benchmark_v2/**/*.py" = ["S603", "S607"]
 
 [tool.ruff.lint.isort]
 lines-after-imports = 2

src/transformers/cli/add_new_model_like.py

@@ -585,24 +585,28 @@ def _add_new_model_like_internal(
 
     # 9. Run linters
     model_init_file = repo_path / "src" / "transformers" / "models" / "__init__.py"
-    subprocess.run(
-        ["ruff", "check", new_module_folder, tests_folder, model_init_file, "--fix"],
+    subprocess.run(  # noqa: S603
+        ["ruff", "check", new_module_folder, tests_folder, model_init_file, "--fix"],  # noqa: S607
         cwd=repo_path,
         stdout=subprocess.DEVNULL,
     )
-    subprocess.run(
-        ["ruff", "format", new_module_folder, tests_folder, model_init_file],
+    subprocess.run(  # noqa: S603
+        ["ruff", "format", new_module_folder, tests_folder, model_init_file],  # noqa: S607
         cwd=repo_path,
         stdout=subprocess.DEVNULL,
     )
-    subprocess.run(
-        ["python", "utils/check_doc_toc.py", "--fix_and_overwrite"], cwd=repo_path, stdout=subprocess.DEVNULL
+    subprocess.run(  # noqa: S603
+        ["python", "utils/check_doc_toc.py", "--fix_and_overwrite"],  # noqa: S607
+        cwd=repo_path,
+        stdout=subprocess.DEVNULL,
     )
-    subprocess.run(["python", "utils/sort_auto_mappings.py"], cwd=repo_path, stdout=subprocess.DEVNULL)
+    subprocess.run(["python", "utils/sort_auto_mappings.py"], cwd=repo_path, stdout=subprocess.DEVNULL)  # noqa: S603, S607
 
     # 10. Run the modular conversion
-    subprocess.run(
-        ["python", "utils/modular_model_converter.py", new_lowercase_name], cwd=repo_path, stdout=subprocess.DEVNULL
+    subprocess.run(  # noqa: S603
+        ["python", "utils/modular_model_converter.py", new_lowercase_name],  # noqa: S607
+        cwd=repo_path,
+        stdout=subprocess.DEVNULL,
     )
 
 

src/transformers/models/auto/processing_auto.py

@@ -442,6 +442,9 @@ def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
                     pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
                 )
             except Exception:
+                logger.debug(
+                    "Failed to load %s with %s, trying next class.", pretrained_model_name_or_path, klass.__name__
+                )
                 continue
 
         raise ValueError(

src/transformers/models/ernie4_5/convert_ernie4_5_tokenizer.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import argparse
+import tempfile
 
 from transformers import LlamaTokenizer, LlamaTokenizerFast
 
@@ -67,6 +68,7 @@
         hf_tok.add_tokens([token], special_tokens=True)
 
     # save slow model and convert on load time
-    hf_tok.save_pretrained("/tmp/ernie4_5_tokenizer")
-    hf_tok_fast = LlamaTokenizerFast.from_pretrained("/tmp/ernie4_5_tokenizer", from_slow=True)
+    with tempfile.TemporaryDirectory() as tmp_dir:
+        hf_tok.save_pretrained(tmp_dir)
+        hf_tok_fast = LlamaTokenizerFast.from_pretrained(tmp_dir, from_slow=True)
     hf_tok_fast.save_pretrained(args.output_dir, push_to_hub=args.push_to_hub)

src/transformers/models/ernie4_5_vl_moe/convert_ernie4_5_vl_moe_to_hf.py

@@ -16,6 +16,7 @@
 import argparse
 import json
 import os
+import tempfile
 from pathlib import Path
 from shutil import copyfile
 
@@ -72,7 +73,7 @@
     "rope_parameters",
 ]
 
-TMP_TOKENIZER_DIR = "/tmp/ernie_vl_tokenizer"
+TMP_TOKENIZER_DIR = tempfile.mkdtemp(prefix="ernie_vl_tokenizer_")
 TOKENIZER_CONFIG_FILE = "tokenizer_config.json"
 DEFAULT_CHAT_TEMPLATE = """
 {%- set image_count = namespace(value=0) -%}

src/transformers/models/hubert/convert_distilhubert_original_s3prl_checkpoint_to_pytorch.py

@@ -14,6 +14,7 @@
 """Convert Hubert checkpoint."""
 
 import argparse
+import ast
 
 import torch
 from s3prl.hub import distilhubert
@@ -158,7 +159,7 @@ def convert_config(model):
     config.apply_spec_augment = False
     config.attention_dropout = fs_config.attention_dropout
     config.conv_bias = False
-    conv_layers = eval(fs_config.extractor_conv_feature_layers)
+    conv_layers = ast.literal_eval(fs_config.extractor_conv_feature_layers)
     config.conv_dim = [x[0] for x in conv_layers]
     config.conv_kernel = [x[1] for x in conv_layers]
     config.conv_stride = [x[2] for x in conv_layers]

src/transformers/models/sew/convert_sew_original_pytorch_checkpoint_to_pytorch.py

@@ -14,6 +14,7 @@
 """Convert SEW checkpoint."""
 
 import argparse
+import ast
 import json
 import os
 
@@ -177,7 +178,7 @@ def convert_config(model, is_finetuned):
         fs_config = model.cfg
 
     config.conv_bias = fs_config.conv_bias
-    conv_layers = eval(fs_config.conv_feature_layers)
+    conv_layers = ast.literal_eval(fs_config.conv_feature_layers)
     config.conv_dim = [x[0] for x in conv_layers]
     config.conv_kernel = [x[1] for x in conv_layers]
     config.conv_stride = [x[2] for x in conv_layers]

src/transformers/models/sew_d/convert_sew_d_original_pytorch_checkpoint_to_pytorch.py

@@ -14,6 +14,7 @@
 """Convert SEW checkpoint."""
 
 import argparse
+import ast
 import json
 import os
 
@@ -181,7 +182,7 @@ def convert_config(model, is_finetuned):
         fs_config = model.cfg
 
     config.conv_bias = fs_config.conv_bias
-    conv_layers = eval(fs_config.conv_feature_layers)
+    conv_layers = ast.literal_eval(fs_config.conv_feature_layers)
     config.conv_dim = [x[0] for x in conv_layers]
     config.conv_kernel = [x[1] for x in conv_layers]
     config.conv_stride = [x[2] for x in conv_layers]

src/transformers/models/videomt/convert_videomt_to_hf.py

@@ -478,8 +478,8 @@ def verify_conversion_against_github_reference(
     with tempfile.TemporaryDirectory(prefix="videomt_ref_") as tmp_dir:
         repo_path = Path(reference_repo_path) if reference_repo_path is not None else Path(tmp_dir) / "videomt"
         if reference_repo_path is None:
-            subprocess.run(
-                ["git", "clone", "--depth", "1", "https://github.com/tue-mps/videomt", str(repo_path)],
+            subprocess.run(  # noqa: S603
+                ["git", "clone", "--depth", "1", "https://github.com/tue-mps/videomt", str(repo_path)],  # noqa: S607
                 check=True,
                 capture_output=True,
                 text=True,

src/transformers/pipelines/audio_classification.py

@@ -51,7 +51,7 @@ def ffmpeg_read(bpayload: bytes, sampling_rate: int) -> np.ndarray:
     ]
 
     try:
-        ffmpeg_process = subprocess.Popen(ffmpeg_command, stdin=subprocess.PIPE, stdout=subprocess.PIPE)
+        ffmpeg_process = subprocess.Popen(ffmpeg_command, stdin=subprocess.PIPE, stdout=subprocess.PIPE)  # noqa: S603
     except FileNotFoundError:
         raise ValueError("ffmpeg was not found but is required to load audio files from filename")
     output_stream = ffmpeg_process.communicate(bpayload)

src/transformers/pipelines/audio_utils.py

@@ -30,7 +30,7 @@ def ffmpeg_read(bpayload: bytes, sampling_rate: int) -> np.ndarray:
     ]
 
     try:
-        with subprocess.Popen(ffmpeg_command, stdin=subprocess.PIPE, stdout=subprocess.PIPE) as ffmpeg_process:
+        with subprocess.Popen(ffmpeg_command, stdin=subprocess.PIPE, stdout=subprocess.PIPE) as ffmpeg_process:  # noqa: S603
             output_stream = ffmpeg_process.communicate(bpayload)
     except FileNotFoundError as error:
         raise ValueError("ffmpeg was not found but is required to load audio files from filename") from error
@@ -265,7 +265,7 @@ def _ffmpeg_stream(ffmpeg_command, buflen: int):
     """
     bufsize = 2**24  # 16Mo
     try:
-        with subprocess.Popen(ffmpeg_command, stdout=subprocess.PIPE, bufsize=bufsize) as ffmpeg_process:
+        with subprocess.Popen(ffmpeg_command, stdout=subprocess.PIPE, bufsize=bufsize) as ffmpeg_process:  # noqa: S603
             while True:
                 raw = ffmpeg_process.stdout.read(buflen)
                 if raw == b"":
@@ -282,7 +282,7 @@ def _get_microphone_name():
     command = ["ffmpeg", "-list_devices", "true", "-f", "dshow", "-i", ""]
 
     try:
-        ffmpeg_devices = subprocess.run(command, text=True, stderr=subprocess.PIPE, encoding="utf-8")
+        ffmpeg_devices = subprocess.run(command, text=True, stderr=subprocess.PIPE, encoding="utf-8")  # noqa: S603
         microphone_lines = [line for line in ffmpeg_devices.stderr.splitlines() if "(audio)" in line]
 
         if microphone_lines:

src/transformers/testing_utils.py

@@ -2544,7 +2544,7 @@ def run_command(command: list[str], return_stdout=False):
     if an error occurred while running `command`
     """
     try:
-        output = subprocess.check_output(command, stderr=subprocess.STDOUT)
+        output = subprocess.check_output(command, stderr=subprocess.STDOUT)  # noqa: S603
         if return_stdout:
             if hasattr(output, "decode"):
                 output = output.decode("utf-8")
@@ -2774,7 +2774,7 @@ def wrapper(*args, **kwargs):
                 else:
                     command = [f"{sys.executable}", "-m", "pytest", f"{test}"]
 
-                subprocess.run(command, env=env, check=True, capture_output=True)
+                subprocess.run(command, env=env, check=True, capture_output=True)  # noqa: S603
             except subprocess.CalledProcessError as e:
                 exception_message = e.stdout.decode()
                 lines = exception_message.split("\n")
@@ -3799,7 +3799,7 @@ def torchrun(script: str, nproc_per_node: int, is_torchrun: bool = True, env: di
 
         # Note that the subprocess will be waited for here, and raise an error if not successful
         try:
-            _ = subprocess.run(cmd, capture_output=True, env=env, text=True, check=True)
+            _ = subprocess.run(cmd, capture_output=True, env=env, text=True, check=True)  # noqa: S603
         except subprocess.CalledProcessError as e:
             raise Exception(f"The following error was captured: {e.stderr}")
 

src/transformers/tokenization_utils_base.py

@@ -3402,6 +3402,7 @@ def find_sentencepiece_model_file(pretrained_model_name_or_path, **kwargs):
                 return candidate
         except Exception:
             # TODO: tighten to OSError / ProxyError
+            logger.debug("Failed to check for %s in %s, skipping.", candidate, pretrained_model_name_or_path)
             continue
 
     subfolder = kwargs.get("subfolder", "")

src/transformers/utils/import_utils.py

@@ -930,7 +930,7 @@ def is_ninja_available() -> bool:
     [ninja](https://ninja-build.org/) build system is available on the system, `False` otherwise.
     """
     try:
-        subprocess.check_output(["ninja", "--version"])
+        subprocess.check_output(["ninja", "--version"])  # noqa: S603, S607
     except Exception:
         return False
     else:
@@ -2407,6 +2407,9 @@ def call(self, *args, **kwargs):
                                     break
                                 except Exception:
                                     # If this candidate fails, try the next one
+                                    logger.debug(
+                                        "Failed to resolve %s via candidate %s, trying next.", name, candidate_name
+                                    )
                                     continue
                             else:
                                 # Candidate not in _class_to_module - might need recursive resolution
@@ -2423,6 +2426,11 @@ def call(self, *args, **kwargs):
                                         setattr(self, name, value)
                                         break
                                 except Exception:
+                                    logger.debug(
+                                        "Failed to resolve %s via recursive lookup of %s, trying next.",
+                                        name,
+                                        candidate_name,
+                                    )
                                     continue
                 except (ImportError, AttributeError):
                     pass

tests/models/dia/test_modeling_dia.py

@@ -14,7 +14,8 @@
 """Testing suite for the PyTorch Dia model."""
 
 import copy
-import pathlib
+import os
+import shutil
 import tempfile
 import unittest
 
@@ -546,14 +547,14 @@ def setUp(self):
         audio_sample_2 = librispeech_dummy[-2]["audio"]["array"]
         # 10 and 5 codebooks as prefix - saved as files as we need wav files for the original Dia
         dac_chunk_len = 512
-        self.audio_prompt_1_path = "/tmp/dia_test_sample_1.mp3"
-        self.audio_prompt_2_path = "/tmp/dia_test_sample_2.mp3"
+        self._tmp_dir = tempfile.mkdtemp(prefix="dia_test_")
+        self.audio_prompt_1_path = os.path.join(self._tmp_dir, "dia_test_sample_1.mp3")
+        self.audio_prompt_2_path = os.path.join(self._tmp_dir, "dia_test_sample_2.mp3")
         sf.write(self.audio_prompt_1_path, audio_sample_1[: (dac_chunk_len * 10)], self.sampling_rate)
         sf.write(self.audio_prompt_2_path, audio_sample_2[: (dac_chunk_len * 5)], self.sampling_rate)
 
     def tearDown(self):
-        pathlib.Path(self.audio_prompt_1_path).unlink()
-        pathlib.Path(self.audio_prompt_2_path).unlink()
+        shutil.rmtree(self._tmp_dir, ignore_errors=True)
         cleanup(torch_device, gc_collect=True)
 
     @slow

tests/tokenization/test_tokenization_utils.py

@@ -345,9 +345,10 @@ def test_special_tokens_overwrite(self):
             == text_with_nonspecial_tokens
         )
 
-        # Checking if this carries over even after saving and relaoding
-        tokenizer.save_pretrained("/tmp/ernie_tokenizer")
-        new_tokenizer = AutoTokenizer.from_pretrained("/tmp/ernie_tokenizer")
+        # Checking if this carries over even after saving and reloading
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            tokenizer.save_pretrained(tmp_dir)
+            new_tokenizer = AutoTokenizer.from_pretrained(tmp_dir)
         self.assertTrue(
             new_tokenizer.decode(new_tokenizer.encode(text_with_nonspecial_tokens), skip_special_tokens=True)
             == text_with_nonspecial_tokens

tests/utils/test_auto_docstring.py

@@ -16,6 +16,7 @@
 """
 
 import importlib
+import logging
 import os
 import statistics
 import sys
@@ -724,6 +725,7 @@ def _import_all():
                 try:
                     importlib.import_module(mod)
                 except Exception:
+                    logging.debug("Failed to import module %s, skipping.", mod)
                     continue
 
         _import_all()  # warmup

utils/add_dates.py

@@ -93,13 +93,13 @@ def check_file_exists_on_github(file_path: str) -> bool:
 def get_modified_cards() -> list[str]:
     """Get the list of model names from modified files in docs/source/en/model_doc/"""
 
-    current_branch = subprocess.check_output(["git", "branch", "--show-current"], text=True).strip()
+    current_branch = subprocess.check_output(["git", "branch", "--show-current"], text=True).strip()  # noqa: S607
     if current_branch == "main":
         # On main branch, only uncommitted changes detected
-        result = subprocess.check_output(["git", "diff", "--name-only", "HEAD"], text=True)
+        result = subprocess.check_output(["git", "diff", "--name-only", "HEAD"], text=True)  # noqa: S607
     else:
-        fork_point_sha = subprocess.check_output("git merge-base main HEAD".split()).decode("utf-8")
-        result = subprocess.check_output(f"git diff --name-only {fork_point_sha}".split()).decode("utf-8")
+        fork_point_sha = subprocess.check_output("git merge-base main HEAD".split()).decode("utf-8")  # noqa: S603
+        result = subprocess.check_output(f"git diff --name-only {fork_point_sha}".split()).decode("utf-8")  # noqa: S603
 
     model_names = []
     for line in result.strip().split("\n"):
@@ -159,8 +159,9 @@ def get_first_commit_date(model_name: str | None) -> str:
         final_date = date.today().isoformat()
     else:
         # File exists on GitHub main branch, get the first commit date from local git history
-        final_date = subprocess.check_output(
-            ["git", "log", "--reverse", "--pretty=format:%ad", "--date=iso", file_path], text=True
+        final_date = subprocess.check_output(  # noqa: S603
+            ["git", "log", "--reverse", "--pretty=format:%ad", "--date=iso", file_path],  # noqa: S607
+            text=True,
         )
     return final_date.strip().split("\n")[0][:10]
 
@@ -208,6 +209,7 @@ def replace_paper_links(file_path: str) -> bool:
 
         except Exception:
             # Paper not available on huggingface, keep arxiv link
+            logger.debug("Paper %s not available on huggingface, keeping arxiv link.", paper_id)
             continue
 
     # Write back only if content changed