[Feature] Spec V2 DFlash Support (#23000)

2026-06-10 19:27:42 -07:00
parent ce0ff154a5
commit 588d1f7bc9
21 changed files with 2439 additions and 250 deletions
@@ -36,6 +36,10 @@ class TestPCGWithDFlash(PCGSpecBase, CustomTestCase):
        "1",
        "--max-running-requests",
        "64",
+        # Keep headroom for the draft KV pool + piecewise cuda graph
+        # private pools on 32GB CI cards.
+        "--mem-fraction-static",
+        "0.7",
        "--cuda-graph-bs-decode",
        *[str(i) for i in range(1, 65)],
    ]