unsloth/studio/backend/tests/test_yaml_trust_remote_code_removed.py
Daniel Han 1582d2854c
Harden trust_remote_code consent: scan GGUF-only auto_map and drop pre-set TRC defaults (#6478)
* Scan auto_map for GGUF-only repo ids in the consent gate

The trust_remote_code consent gate treated any repo classified GGUF-only
(ships .gguf, no transformers-loadable weight) as having no remote code,
so _config_has_auto_map returned False even when a config declared an
auto_map and the repo shipped the referenced .py. The evaluator then
skipped the scan/fingerprint for that target entirely.

GGUF-inertness is a property of the loader, not the repo. A GGUF
selection loads via llama.cpp, which never reads config.json/auto_map,
and that case is already short-circuited upstream by the caller's
is_gguf check (the inference route skips the remote-code preflight for a
GGUF load). Every path that reaches this helper (export, training,
non-GGUF inference) loads through transformers/Unsloth from_pretrained,
which DOES import auto_map even for a repo that only ships .gguf weights:
the custom module runs before from_pretrained fails on the missing
transformers weights. The export path has no is_gguf guard and passes the
source straight to FastLanguageModel.from_pretrained(trust_remote_code=True),
so the in-helper GGUF skip let a repo with config.json (auto_map) +
modeling_x.py + only a .gguf run unreviewed code during export.

Drop the redundant repo-level GGUF short-circuit (and the now-unused
_is_gguf_repo helper). A direct .gguf file reference stays inert via
_is_direct_gguf_file_ref because that genuinely is a single-file llama.cpp
load; repo ids are always scanned. A GGUF repo whose auto_map ships no .py
still allows via the existing empty-code path, so legitimate GGUF loads
are unaffected (and GGUF inference never reaches this helper at all). Only
a repo that actually contains a .gguf can change behavior here; non-GGUF
repos (safetensors, MLX) are byte-identical before and after.

Update the GGUF auto_map test to expect a scan, and add two regression
tests: a GGUF-only repo shipping auto_map Python is scanned and blocked,
and a transformers-style repo (safetensors / MLX .npz) with auto_map stays
scanned and blocked.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Remove trust_remote_code config defaults; consent dialog is the only enabler

trust_remote_code is a per-load decision that must go through the remote-code
consent dialog, which scans the auto_map code and pins the exact version. Two
pre-set paths could still enable it without the user reviewing any code, and the
GGUF consent bypass rode one of them into the export flow:

- 4 model_defaults YAMLs shipped trust_remote_code: true (GLM-4.7-Flash,
  Nemotron-3-Nano-30B-A3B, PaddleOCR-VL, ERNIE-4.5-VL).
- The frontend consent hook silently enabled trust_remote_code on a clean scan
  whenever the caller flagged the model as needing it.

Remove every trust_remote_code key from the model_defaults YAMLs (the loaders
already default to False when the key is absent) and delete the frontend silent
auto-enable, so trust_remote_code is only turned on after the user approves the
scanned code in the dialog.

The three models that genuinely run custom code ship auto_map, which the consent
gate detects on its own via _config_has_auto_map, so the dialog still fires for
them in inference, training, and export (Nemotron is also re-granted by the
trusted-org auto-enable in the workers). GLM-4.7-Flash has no auto_map:
glm4_moe_lite is native in transformers 5.0+ and it loads with
trust_remote_code=False, so its YAML flag was a no-op.

Adds test_yaml_trust_remote_code_removed.py.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Drop YAML sections emptied by trust_remote_code removal

Removing trust_remote_code from a model YAML whose section had no other key
left a bare `inference:` header, which PyYAML parses as None;
load_inference_config() then does `model_config.get("inference", {}).get(...)`
and crashes on the None. Drop those now-empty section headers (24 model
defaults, all the `inference:` section) so callers fall back to family/default
inference params, which is the same result those models had before (their only
inference override was trust_remote_code).

Strengthens test_yaml_trust_remote_code_removed.py to forbid any empty/None
top-level section and to load the affected models' inference config end to end.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Add sweep asserting every model YAML loads via training + inference paths

Loads all model_defaults YAMLs through load_model_defaults (training) and
load_inference_config (inference) with the exact .get() access patterns the
routes use, so a malformed/None section that crashes either loader is caught.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Assert ex-TRC auto_map models still surface the consent dialog

Removing the trust_remote_code YAML default must not suppress the dialog for the
models that genuinely run custom code. The dialog is driven by the repo's auto_map
(via preflight_remote_code_consent_for_targets -> _config_has_auto_map), not the YAML
flag, so Nemotron/PaddleOCR-VL/ERNIE-4.5-VL still require consent; GLM-4.7-Flash (no
auto_map) takes no dialog and loads natively. Mocks only the Hub config + .py reader.

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci

* Tighten comments in consent-gate changes

* Trim comments to be more succinct

---------

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Daniel Han <michaelhan2050@gmail.com>
2026-06-22 02:10:35 -07:00

163 lines
6.8 KiB
Python

# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
"""Regression: model-default YAMLs must not pre-set trust_remote_code.
It is a per-load decision made through the consent dialog (which scans and pins the
auto_map code), never a config default -- a YAML flag would re-open the no-review
bypass. Models that run custom code ship auto_map, so the dialog still fires without it.
"""
from pathlib import Path
import yaml
_CONFIGS = Path(__file__).resolve().parent.parent / "assets" / "configs"
_MODEL_DEFAULTS = _CONFIGS / "model_defaults"
def test_no_model_default_yaml_sets_trust_remote_code():
offenders = []
for f in _MODEL_DEFAULTS.rglob("*.yaml"):
doc = yaml.safe_load(f.read_text()) or {}
if not isinstance(doc, dict):
continue
for section, body in doc.items():
if isinstance(body, dict) and "trust_remote_code" in body:
offenders.append(
f"{f.relative_to(_CONFIGS)} [{section}={body['trust_remote_code']}]"
)
assert not offenders, (
"trust_remote_code must not be pre-set in model defaults; it is enabled only via "
f"the consent dialog. Remove it from: {offenders}"
)
def test_no_model_default_yaml_has_empty_or_none_section():
# A bare `inference:` header (no keys) parses to None and crashes the .get() loaders.
offenders = []
for f in _MODEL_DEFAULTS.rglob("*.yaml"):
doc = yaml.safe_load(f.read_text())
if not isinstance(doc, dict):
offenders.append(f"{f.relative_to(_CONFIGS)} (not a mapping)")
continue
for section, body in doc.items():
if body is None or (isinstance(body, dict) and not body):
offenders.append(f"{f.relative_to(_CONFIGS)} [{section}]")
assert not offenders, (
"empty/None YAML section would crash the config loaders; drop the bare section "
f"header instead. Offending: {offenders}"
)
def test_formerly_flagged_models_load_inference_config_without_crash():
# Models whose inference section was emptied by the TRC removal must still load.
from utils.inference import load_inference_config
for model in (
"tiiuae/Falcon-H1-0.5B-Instruct",
"unsloth/Llama-3.2-1B-Instruct",
"unsloth/Qwen2.5-7B",
):
cfg = load_inference_config(model)
assert isinstance(cfg, dict)
assert cfg.get("trust_remote_code", False) is False
def test_all_model_yamls_load_for_training_and_inference():
# Every YAML must load through both config paths (training + inference) as the routes do.
from utils.inference import load_inference_config
from utils.models.model_config import load_model_defaults
infer_keys = {
"temperature",
"top_p",
"top_k",
"min_p",
"presence_penalty",
"trust_remote_code",
}
failures = []
for f in sorted(_MODEL_DEFAULTS.rglob("*.yaml")):
stem = f.stem
try:
md = load_model_defaults(stem)
assert isinstance(md, dict), f"load_model_defaults -> {type(md).__name__}"
assert not [k for k, v in md.items() if v is None], "has a None section"
# the dict sections the loaders read via .get('sect', {}).get(...)
for sect in ("training", "inference", "lora", "logging"):
assert isinstance(md.get(sect, {}), dict), f"{sect!r} is not a mapping"
md.get("training", {}).get("trust_remote_code", False) # routes/training.py:263
cfg = load_inference_config(stem)
assert infer_keys <= set(cfg), f"inference config missing {infer_keys - set(cfg)}"
except Exception as e: # noqa: BLE001 - aggregate so one failure does not hide others
failures.append(f"{f.relative_to(_CONFIGS)}: {type(e).__name__}: {e}")
assert not failures, "YAML config loaders crashed on: " + "; ".join(failures)
def test_base_templates_have_no_trust_remote_code():
for name in ("full_finetune.yaml", "lora_text.yaml", "vision_lora.yaml"):
doc = yaml.safe_load((_CONFIGS / name).read_text()) or {}
flat = yaml.safe_dump(doc)
assert "trust_remote_code" not in flat, f"{name} should not set trust_remote_code"
def test_loader_defaults_trust_remote_code_off_for_formerly_flagged_models():
# The 4 models that used to ship trust_remote_code: true must now report no default.
from utils.models.model_config import load_model_defaults
for model in (
"unsloth/GLM-4.7-Flash",
"unsloth/Nemotron-3-Nano-30B-A3B",
"unsloth/PaddleOCR-VL",
"unsloth/ERNIE-4.5-VL-28B-A3B-PT",
):
d = load_model_defaults(model)
for section in ("training", "inference"):
assert not (d.get(section) or {}).get(
"trust_remote_code", False
), f"{model} [{section}] still carries a trust_remote_code default"
def test_formerly_flagged_auto_map_models_still_require_consent_dialog():
# Crux: an auto_map model must STILL surface the dialog (driven by auto_map, not the
# YAML flag). Real backend path, mocking only the Hub json + .py fetch.
from unittest.mock import patch
from utils.security import consent, preflight_remote_code_consent_for_targets
auto_map_cfg = [
{
"auto_map": {
"AutoConfig": "configuration_x.XConfig",
"AutoModelForCausalLM": "modeling_x.XForCausalLM",
}
}
]
benign_py = {"modeling_x.py": "class XForCausalLM:\n pass\n"}
for model in (
"unsloth/Nemotron-3-Nano-30B-A3B",
"unsloth/PaddleOCR-VL",
"unsloth/ERNIE-4.5-VL-28B-A3B-PT",
):
with (
patch.object(consent, "_load_remote_code_configs", return_value = auto_map_cfg),
patch.object(consent, "repo_remote_code_files", return_value = benign_py),
):
decision = preflight_remote_code_consent_for_targets([model], hf_token = None)
# routes/models.py opens the dialog from decision.has_remote_code.
assert decision.has_remote_code is True, (
f"{model} ships auto_map but the consent scan did not flag it -> dialog would "
"not fire"
)
def test_no_auto_map_model_takes_no_dialog():
# Flip side: GLM-4.7-Flash ships no auto_map -> no dialog; its old YAML flag was a no-op.
from unittest.mock import patch
from utils.security import consent, preflight_remote_code_consent_for_targets
with patch.object(
consent, "_load_remote_code_configs", return_value = [{"model_type": "glm4_moe_lite"}]
):
decision = preflight_remote_code_consent_for_targets(
["unsloth/GLM-4.7-Flash"], hf_token = None
)
assert decision.has_remote_code is False