[Kernel] Migrate generic attention kernels to sglang.kernels (RFC #29630, Phase 2.5, 4/7) (#30789)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-14 16:53:46 +08:00
parent a5a71c6c26
commit 1a35440c4a
35 changed files with 173 additions and 143 deletions
@@ -1837,7 +1837,7 @@ class MHATokenToKVPool(KVCache):
        # and viewed as ``store_dtype`` by ``set_kv_buffer``.
        if self.kv_cache_layout == "vectorized_5d":
            # Late-import to keep the NHD path import-clean.
-            from sglang.srt.layers.attention.utils import (
+            from sglang.kernels.ops.attention.utils import (
                launch_reshape_and_cache_shuffle_5d,
            )