ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

respect trust_remote_code toggle, return helpful error when required

Browse source
This commit is contained in:
Roland Tannous 2026-03-09 13:06:55 +00:00
commit 7989cd4567
3 changed files with 28 additions and 23 deletions
Download patch file Download diff file Expand all files Collapse all files

16
studio/backend/core/training/worker.py
View file

@ -198,9 +198,23 @@ def run_training_process(
if trainer.should_stop:
event_queue.put({"type": "complete", "output_dir": None, "ts": time.time()})
else:
error_msg = trainer.training_progress.error or "Failed to load model"
# Hint about trust_remote_code if YAML says this model needs it
if not config.get("trust_remote_code", False):
try:
from utils.models.model_config import load_model_defaults
model_defaults = load_model_defaults(model_name)
yaml_trust = model_defaults.get("training", {}).get("trust_remote_code", False)
if yaml_trust:
error_msg = (
f"Model '{model_name}' requires trust_remote_code to be enabled. "
f"Please enable 'Trust remote code' in Chat Settings and try again."
)
except Exception:
pass
event_queue.put({
"type": "error",
"error": trainer.training_progress.error or "Failed to load model",
"error": error_msg,
"stack": "", "ts": time.time(),
})
return

25
studio/backend/routes/inference.py
View file

@ -222,27 +222,28 @@ async def load_model(
except Exception as e:
logger.warning(f"Could not read adapter_config.json: {e}")
# Resolve trust_remote_code: use True if either the request or YAML config says so.
# This ensures models like Nemotron that require it always get it, even if the
# frontend toggle hasn't been set yet.
trust_remote_code = request.trust_remote_code
if not trust_remote_code:
model_defaults = load_model_defaults(config.identifier)
yaml_trust = model_defaults.get("inference", {}).get("trust_remote_code", False)
if yaml_trust:
logger.info(f"YAML config sets trust_remote_code=True for {config.identifier}")
trust_remote_code = True
# Load the model
success = backend.load_model(
config=config,
max_seq_length=request.max_seq_length,
load_in_4bit=load_in_4bit,
hf_token=request.hf_token,
trust_remote_code=trust_remote_code,
trust_remote_code=request.trust_remote_code,
)
if not success:
# Check if YAML says this model needs trust_remote_code
if not request.trust_remote_code:
model_defaults = load_model_defaults(config.identifier)
yaml_trust = model_defaults.get("inference", {}).get("trust_remote_code", False)
if yaml_trust:
raise HTTPException(
status_code=400,
detail=(
f"Model '{config.display_name}' requires trust_remote_code to be enabled. "
f"Please enable 'Trust remote code' in Chat Settings and try again."
),
)
raise HTTPException(
status_code=500,
detail=f"Failed to load model: {config.display_name}"

10
studio/backend/routes/training.py
View file

@ -19,14 +19,12 @@ if str(backend_path) not in sys.path:
# Import backend functions
try:
from core.training import get_training_backend
from utils.models.model_config import load_model_defaults
except ImportError:
# Fallback: try to import from parent directory
parent_backend = backend_path.parent / "backend"
if str(parent_backend) not in sys.path:
sys.path.insert(0, str(parent_backend))
from core.training import get_training_backend
from utils.models.model_config import load_model_defaults
# Auth
from auth.authentication import get_current_subject
@ -196,14 +194,6 @@ async def start_training(
"trust_remote_code": request.trust_remote_code,
}
# Resolve trust_remote_code: use True if either the request or YAML config says so.
if not training_kwargs["trust_remote_code"]:
model_defaults = load_model_defaults(request.model_name)
yaml_trust = model_defaults.get("training", {}).get("trust_remote_code", False)
if yaml_trust:
logger.info(f"YAML config sets trust_remote_code=True for {request.model_name}")
training_kwargs["trust_remote_code"] = True
# Free GPU memory: shut down any running inference/export subprocesses
# before training starts (they'd compete for VRAM otherwise)
try: