This repository contains a minimal NPU-only inference entrypoint for the
published PaulVialard/ppo-Huggy ML-Agents policy. It selects the latest
Huggy-*.pt checkpoint from the fixed upstream revision, reconstructs the
official 59→512→512→512→21 continuous-action network, and executes the policy
forward pass on Ascend NPU.
The model snapshot is not redistributed. The back-half platform is CANN 8.5.1,
PyTorch 2.9.0, and torch_npu 2.9.0.post1. Keep that platform stack unchanged
and restore the exact official GitCode model revision:
git clone https://gitcode.com/hf_mirrors/PaulVialard/ppo-Huggy.git model
git -C model checkout --detach 1fd517574882e4b765aa24a2b078d91fa1ab9072
git -C model lfs pull
python3 -m pip install --target ./pydeps -r requirements.txt
source /usr/local/Ascend/ascend-toolkit/set_env.sh
PYTHONPATH="$PWD/pydeps" python inference.py --model-dir ./model --device npu:0 \
--samples 8 --output result.jsonThe JSON result records every device in the model/input/output chain, output shapes and counts, finite/non-finite counts, timing, and exit status. The script has no CPU/CUDA bridge or fallback; host work is limited to loading the checkpoint and serializing the audit result.
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.


The device views show the same Python PID at two distinct times. The result views show complete and concise views of the same successful real-NPU run.




本仓为 2301_76761127 队伍交付件(与备份账号合并提交)。