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

backend: resolve trust_remote_code from YAML when not set by frontend

Browse source
This commit is contained in:
Roland Tannous 2026-03-09 11:58:23 +00:00
commit 4858204c62
2 changed files with 24 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

15
studio/backend/routes/inference.py
View file

@ -26,6 +26,7 @@ try:
from core.inference.llama_cpp import LlamaCppBackend
from utils.models import ModelConfig
from utils.inference import load_inference_config
from utils.models.model_config import load_model_defaults
except ImportError:
parent_backend = backend_path.parent / "backend"
if str(parent_backend) not in sys.path:
@ -34,6 +35,7 @@ except ImportError:
from core.inference.llama_cpp import LlamaCppBackend
from utils.models import ModelConfig
from utils.inference import load_inference_config
from utils.models.model_config import load_model_defaults
from models.inference import (
LoadRequest,
@ -220,13 +222,24 @@ 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=request.trust_remote_code,
trust_remote_code=trust_remote_code,
)
if not success:

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

@ -19,12 +19,14 @@ 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
@ -194,6 +196,14 @@ 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: