Init attention backend for Intel XPU (#10656)

Co-authored-by: guangyey <guangye.yu@intel.com> Co-authored-by: DiweiSun <105627594+DiweiSun@users.noreply.github.com>
2025-10-21 11:41:28 +08:00
parent fb6cc7b000
commit b113c72e7a
18 changed files with 1210 additions and 26 deletions
@@ -272,7 +272,7 @@ def prepare_synthetic_inputs_for_latency_test(
 def extend(reqs, model_runner):
    # Create dummy tree_cache for benchmarks (no prefix caching, just allocation)
    dummy_tree_cache = SimpleNamespace(
-        page_size=1,
+        page_size=model_runner.server_args.page_size,
        device=model_runner.device,
        token_to_kv_pool_allocator=model_runner.token_to_kv_pool_allocator,
    )