fix(pd): transfer MiniMax-M3 sparse indexer-key cache in disaggregation

[Jasen2201]

MiniMax-M3 sparse attention reuses the unified KV cache and kv_scale for K/V, so the fp8 per-token scales already travel with the KV blocks. It keeps one extra per-token buffer, runner.sparse_attention_index_cache, holding the indexer keys used for top-k block selection at decode time. get_kv_transfer_tensors() never registered that buffer, so under PD disaggregation the decode node ran top-k against a zero/stale index for the prefilled tokens and attended to the wrong KV blocks. This is masked for short prompts (the init+local+topk window already covers every block, so selection is moot) but corrupts output once the context exceeds that window.

Register the indexer-key cache as block-indexed transfer regions (one per sparse layer, same physical-block striding as the KV cache), guarded by getattr so non-sparse models and bf16 paths are unaffected.

Tested (latest image, 1P+1D TP4, fp8 KV via Triton attention): GSM8K 5-shot = 0.9401, i.e. no regression to M3 fp8 PD. Short-prompt GSM8K does not exercise the long-context top-k path the buffer affects; that path is covered by review, not this run.

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

• Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

[Copilot]

Pull request overview

Fixes PD disaggregation correctness for MiniMax-M3 sparse attention by ensuring the per-token sparse indexer-key cache is included in the KV RDMA transfer set. This prevents decode workers from running top‑k block selection against stale/zero index-cache data for prefilled tokens, which can mis-select KV blocks once context grows beyond the init/local/top‑k coverage window.

Changes:

• Register runner.sparse_attention_index_cache as additional block-indexed transfer regions in get_kv_transfer_tensors().
• Guard the new transfer registration via getattr(...) so non-sparse models (and runners without the buffer) are unaffected.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.