This is the sole offline NPU inference entry for theforecastingcompany/t0-alpha, fixed revision f8727c2357e0d81f1d9f56fe3aaac43068b5fc72. Place that exact snapshot at ./model; model weights are not committed here.
The official hosted rerun passed on Ascend 910B3 with Python 3.11, CANN
8.5.1, platform PyTorch 2.9.0 and torch_npu 2.9.0. Keep that platform
pair and rebuild only model-layer dependencies:
git clone https://gitcode.com/2301_78565107/t0-alpha.git repo
git clone --no-checkout https://huggingface.co/theforecastingcompany/t0-alpha model
git -C model checkout f8727c2357e0d81f1d9f56fe3aaac43068b5fc72
git -C model lfs pull
python3 -m pip install --target ./repo/pydeps -r ./repo/requirements.txt
cd repo
source /usr/local/Ascend/ascend-toolkit/set_env.sh
export TORCH_DEVICE_BACKEND_AUTOLOAD=0
PYTHONPATH="$PWD/pydeps" python3 inference.py \
--model-path ../model --device npu:0 --output audit/npu_result.jsonThe model source is gated by its upstream terms, but no private AUX or hidden Python module is required. The entry loads the complete T0Forecaster checkpoint and runs its quantile forecast for an eight-step horizon entirely on NPU.
CPU/CUDA fallback is forbidden. The result records actual model/input/output devices, parameter count, output shape/value count and exit status. Precision and finiteness are recorded honestly but are non-blocking under the current environment-pass policy. Private logs remain outside this public repository.
The two DS-V4-Flash workflow views are independent plan-only conversations: an overall Ascend NPU migration design and a separate dependency, operator, full-NPU validation, and troubleshooting plan. They do not execute tools or modify repository files.


本仓为 2301_76761127 队伍交付件(与备份账号合并提交)。