Download code/tt_moge/reference/__init__.py from changh95/moge-2-p150: direct link, hf CLI and curl.
- Browser
- Download file 573 Bytes
-
https://huggingface.co/changh95/moge-2-p150/resolve/main/code/tt_moge/reference/__init__.py
- Command line
-
hf download hf://changh95/moge-2-p150/code/tt_moge/reference/__init__.py
-
curl -L -o __init__.py https://huggingface.co/changh95/moge-2-p150/resolve/main/code/tt_moge/reference/__init__.py
573 Bytes
| # SPDX-FileCopyrightText: (c) 2026 Tenstorrent AI ULC | |
| # SPDX-License-Identifier: Apache-2.0 | |
| """Torch reference for MoGe-2. | |
| The faithful reference is the upstream `microsoft/MoGe` model (vendored here as | |
| the ``moge`` package, with its ``utils3d`` dependency). Unlike a from-scratch | |
| re-implementation, this guarantees the TT-NN port is graded against the exact | |
| published model. ``load_pretrained.load_moge2`` returns a ready ``MoGeModel``. | |
| """ | |
| import os | |
| import sys | |
| _HERE = os.path.dirname(os.path.abspath(__file__)) | |
| if _HERE not in sys.path: | |
| sys.path.insert(0, _HERE) | |