[chx96642264]

Motivation

PR #30393 generalized the MLA HiCache backup path to support packed target and draft KV buffers. However,  NPUMLATokenToKVPool intentionally does not create data_ptrs (it uses contiguous multi-layer K/V/index-K tensors for the NPU transfer kernel). 
And, Ascend NPU does not support ggml_moe_a8_vec which in sgl_kernel.

Errors as follows：

Modifications

Accuracy Tests

use DeepSeek-V2-Lite-W8A8 test gsm8k

Speed Tests and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

---

CI States

Latest PR Test (Base): ❌ Run #33071938518
Latest PR Test (Extra): ❌ Run #33071938243
Latest PR Test (AMD ROCm 7.2): ❌ Run #33071938450

[sglang-npu-bot]

/tag-and-rerun-ci

[Fridge003]

The root fix is in #36529