| """Hugging Face Inference Endpoints handler (custom container task). |
| |
| Deploy: on the model page choose Deploy -> Inference Endpoints, pick a GPU (or CPU) instance; the |
| endpoint installs requirements.txt and serves this class. Send a Jev/SystemOne request as the JSON |
| body, either directly or under "inputs"; a list of requests returns a list of responses. |
| """ |
| import os |
| import sys |
| from typing import Any |
|
|
| import torch |
|
|
| sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) |
| from decider_model import load_release_model, systemone |
|
|
|
|
| class EndpointHandler: |
| def __init__(self, path: str = ""): |
| path = path or os.path.dirname(os.path.abspath(__file__)) |
| self.model = load_release_model(path, device="cuda" if torch.cuda.is_available() else "cpu") |
|
|
| def __call__(self, data: dict[str, Any]) -> Any: |
| request = data.get("inputs", data) if isinstance(data, dict) else data |
| if isinstance(request, list): |
| return [systemone(self.model, r) for r in request] |
| return systemone(self.model, request) |
|
|
