From 3d34f8cbdcc9545571f5c2ec78e0327b1e5779e4 Mon Sep 17 00:00:00 2001 From: Tuukka Sarvi Date: Tue, 16 Jun 2026 17:28:06 +0300 Subject: [PATCH] [ROCm][Cleanup] Remove stale AITER FA hybrid KV-cache TODO (#44178) Signed-off-by: Tuukka Sarvi Co-authored-by: Cyrus Leung --- vllm/v1/attention/backends/rocm_aiter_fa.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/vllm/v1/attention/backends/rocm_aiter_fa.py b/vllm/v1/attention/backends/rocm_aiter_fa.py index a9fa45debc..7e850af3e7 100644 --- a/vllm/v1/attention/backends/rocm_aiter_fa.py +++ b/vllm/v1/attention/backends/rocm_aiter_fa.py @@ -1408,8 +1408,6 @@ class AiterFlashAttentionImpl(AttentionImpl): assert k_scale is not None and v_scale is not None, ( "k_scale and v_scale are required for shuffled update" ) - # TODO: Add correct KV cache handling for hybrid model. KV cache - # may not be contiguous if mamba state exists. reshape_and_cache_shuffle_triton( key, value,