vllm-ascend-model-adapter

What problem does it solve?

Adapt and debug existing or new models for vLLM on Ascend NPU. Implement in /vllm-workspace/vllm and /vllm-workspace/vllm-ascend, validate via direct vllm serve from /workspace, and deliver one signed commit in the current repo.

Core Features & Use Cases

• Adapt Hugging Face or local models to run on vLLM Ascend with minimal changes.
• Deterministic validation via startup/inference smoke tests, ACLGraph/MTP/EP as applicable.
• Single-commit handoff delivering the final adapter to the target repository.

Quick Start

Run the vLLM Ascend model adapter workflow from /workspace, validate via direct vLLM serve, and deliver a single signed commit to the current repository.