generative-recommendation-verification - Validate generative recommendation models on Ascend NPUs
Automates eight-stage validation of generative recommendation models on Huawei Ascend NPU devices, including environment, training, monitoring, and performance checks.
Tags
Updated: 2026-10-04Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Validate Ascend environments
- Verify dependency versions
- Test compiled operators
- Prepare source and data
- Validate model configurations
- Launch model training
- Monitor training metrics
- Compare training performance
Inputs
- Ascend device path
- Configuration file path
- Working directory
- Model source repositories
- Model dataset
Outputs
- JSON verification report
- Markdown verification report
- Stage pass or failure statuses
- Error logs and diagnostics
- Performance comparison metrics
Requirements
- bash
- python3
- npu-smi
- git
- Ascend 800 A2 or A3
- Ascend 910
- Atlas 800 A2 or A3
- Python 3.11 or later
- torch 2.6.0
- torch_npu 2.6.0
- CANN 8.5.1 or later
- Ascend driver at /usr/local/Ascend/driver