vllm-ascend-model-adapter - Adapt Models for vLLM on Ascend NPU
Adapt Hugging Face or local models to run on vllm-ascend with minimal changes
Tags
Updated: 2026-05-11Capabilities
Typical Inputs
Typical Outputs
What this skill does
- analyze model architecture
- inspect model files
- implement model adapter
- add model processor
- register model support
- map model weights
- validate dummy weights
- validate real weights
- start vllm serve
- send inference request
- verify multimodal features
- generate test config
- create tutorial doc
- backport code changes
- create signed commit
Inputs
- Model path
- Config files
- Processor files
- Modeling files
- Tokenizer files
- Model checkpoint
- vllm-workspace directories
- Current git repository
- Docker environment
Outputs
- Signed git commit
- Test config YAML
- Tutorial markdown
- Tutorial index update
- Analysis report
- Runbook document
- Feature status matrix
- Commit hash
Requirements
- Docker environment
- Git repository
- vllm-workspace installation
- Ascend NPU hardware
- OpenAI-compatible API