Download decision_index_engine.py from chiranko/PKU-KCL-Meteor-2B: direct link, hf CLI and curl.
- Browser
- Download file 2.36 kB
-
https://huggingface.co/chiranko/PKU-KCL-Meteor-2B/resolve/main/decision_index_engine.py
- Command line
-
hf download hf://chiranko/PKU-KCL-Meteor-2B/decision_index_engine.py
-
curl -L -o decision_index_engine.py https://huggingface.co/chiranko/PKU-KCL-Meteor-2B/resolve/main/decision_index_engine.py
2.36 kB
| # Copyright 2026 Knowledge Computing Lab (KCL), Peking University | |
| # SPDX-License-Identifier: Apache-2.0 | |
| """Decision Index engine for PKU-KCL-Meteor, run in process by the kit. | |
| From this repository's directory (or with it on PYTHONPATH): | |
| python -m decision_index pipeline --engine decision_index_engine:MeteorEngine --out runs/<name> | |
| Options (``--option key=value``): ``model`` (this directory by default, or a Hub id), ``revision``, | |
| ``device`` (default ``cuda:0``), ``batch_size`` (at most this many questions per forward pass, default 32), | |
| ``max_batch_tokens`` (at most this many padded tokens per forward pass, default 32768; a longer single | |
| question still runs whole) and ``verify`` (``fast``, ``full`` or ``none`` check of the files against | |
| MODEL_MANIFEST.json, default ``fast``). | |
| """ | |
| import sys | |
| from pathlib import Path | |
| from decision_index.engines.base import Engine, Unsupported | |
| ROOT = Path(__file__).resolve().parent | |
| if str(ROOT) not in sys.path: | |
| sys.path.insert(0, str(ROOT)) | |
| class MeteorEngine(Engine): | |
| name = "pku-kcl-meteor" | |
| latency = ("In-process request wall time: prompt rendering, tokenization, one forward pass per question " | |
| "and the readout; excludes model loading.") | |
| def __init__(self, model=None, revision=None, device="cuda:0", batch_size=32, max_batch_tokens=32768, | |
| verify="fast", **options): | |
| if options: | |
| raise ValueError(f"unknown MeteorEngine options: {sorted(options)}") | |
| super().__init__() | |
| from meteor import Meteor | |
| self.model = Meteor(model, revision=revision, device=device, batch_size=batch_size, | |
| max_batch_tokens=max_batch_tokens, verify=verify) | |
| self.provenance = {"kind": "in-process", **self.model.provenance()} | |
| def __call__(self, state, questions, **inputs): | |
| from meteor import CapacityError | |
| if inputs: | |
| raise Unsupported("PKU-KCL-Meteor reads text only; got " + ", ".join(sorted(inputs))) | |
| try: | |
| response = self.model.decide(state, questions) | |
| except CapacityError as error: | |
| raise Unsupported(str(error)) from None | |
| return response, response | |
| def runtime(self): | |
| return self.model.runtime() | |
| def synchronize(self): | |
| self.model.synchronize() | |
| def close(self): | |
| self.model.close() | |