[jiangkuaixue123]

Summary

• Add the NPU async DP framework plumbing, async CAM connector support, and related runtime/config integration.
• Add async DP patch helpers, offline scheduling utilities, PCP debug helpers, and DeepSeek v2 async MoE support.
• Expand unit coverage for async DP patches, connector metadata, offline schedules, config validation, and NPU worker/runtime behavior.

Impact

This enables the NPU async DP path to be exercised behind the AFD async configuration while keeping the changes isolated in the async connector/runtime flow.

Validation

• Not run in this environment: local python is unavailable and python3 does not have pytest installed.
• tests/unit/v1/worker/test_npu_runtime.py conflict was resolved and staged before the branch was finalized.

[gemini-code-assist]

Code Review

This pull request introduces an asynchronous data-parallel (async-DP) connector (AFDAsyncConnector) along with compatibility patches, offline CSV scheduling, and NPU-specific model runner and worker updates for the Ascend NPU AFD plugin. The code review identified several critical runtime issues that must be addressed, including NameErrors due to missing imports or undefined functions (Mapping and _resolve_graph_pool), potential KeyErrors and TypeErrors during metadata sorting and cloning, negative slicing bugs when slicing token payloads, and a fallback issue where ffn_size is referenced instead of expert_rank_size.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[jiangkuaixue123]

这里的类型可以收窄成 Tensor / Tensor | None。调用点前面已经要求 topk_ids/topk_weights 存在，使用 object + getattr(shape) 的防御性写法有点重，直接访问 .shape 会更清楚。