[FullCG] Support chunked cached-prefix prefill (#30825)

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-28 11:20:25 -07:00
parent 1eee8fbdcc
commit 4e5a05148a
10 changed files with 746 additions and 37 deletions
@@ -20,6 +20,7 @@ class TestPrefillCudaGraphPadding(CustomTestCase):
        runner = PrefillCudaGraphRunner.__new__(PrefillCudaGraphRunner)
        runner._is_full_backend = False
        runner.enable_lora = False
+        runner._capture_chunked_prefix = False
        runner.prefill_backend_name = Backend.TC_PIECEWISE
        runner.has_mha_companion_layers = False
        runner.capture_hidden_mode = CaptureHiddenMode.NULL