aws / aws/sagemaker-python-sdk
ModelBuilder fails w/ DJL container using HF_MODEL_ID
- Lingua principale
- Python
- Stelle
- 2.3k
- Fork
- 1.3k
- Merge medio
- 1g 22h
- PR unite (30g)
- 35
Descrizione
**PySDK Version**
- [ ] PySDK V2 (2.x)
- [X] PySDK V3 (3.4.0)
**Describe the bug**
When using ModelBuilder in "passthrough" mode (image_uri + env_vars, no model
or inference_spec), `build()` fails with a ValidationException because a
non-existent S3 path is injected as `ModelDataUrl` in the CreateModel request.
Passthrough mode is the intended path for deploying models on custom/DJL
containers where the container itself handles model loading (e.g. via
HF_MODEL_ID env var pointing to S3).
**Expected behavior**
Expected: Model created with no ModelDataUrl (container loads via HF_MODEL_ID)
Actual: ValidationException -- "Could not find model data at
s3:///model-builder///"
**Suggested Fix**
In `_build_for_passthrough()`, also clear `s3_model_data_url`:
def _build_for_passthrough(self) -> Model:
if not self.image_uri:
raise ValueError("image_uri is required for pass-through cases")
self.s3_upload_path = None
self.s3_model_data_url = None # <-- add this line
return self._create_model()
**To reproduce**
```
from sagemaker.core.helper.session_helper import Session, get_execution_role
from sagemaker.serve.model_builder import ModelBuilder
sagemaker_session = Session()
region = sagemaker_session.boto_session.region_name
role = get_execution_role(sagemaker_session, use_default=True)
bucket = sagemaker_session.default_bucket()
image_uri = f"763104351884.dkr.ecr.{region}.amazonaws.com/djl-inference:0.33.0-lmi15.0.0-cu128"
model_builder = ModelBuilder(
image_uri=image_uri,
role_arn=role,
sagemaker_session=sagemaker_session,
instance_type="ml.g5.2xlarge",
env_vars={
"HF_MODEL_ID": f"s3://{bucket}/models/any-model",
"OPTION_ROLLING_BATCH": "vllm",
},
)
model_builder.build(model_name="passthrough-bug-repro")
Guida per i contributori
Apri la guida per i contributori
Direzione di ricerca
Inizia da ModelBuilder._build_for_passthrough() e traccia i valori passati a _create_model(). Riproduci la build passthrough con image_uri e HF_MODEL_ID, quindi verifica che una richiesta CreateModel riuscita ometta ModelDataUrl e che il container possa caricare il modello configurato da S3.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Valutazione
- Stack tecnologico
- aws, python
- Ambito
- cloud, machine-learning
- Tipo di issue
- Bug
- Difficoltà
- 2/5
- Tempo stimato
- 1-3 ore
- Stato di attività
- Tranquilla
- Chiarezza
- Specificata chiaramente
- Idoneità per principianti
- 68/100