[DLLM] Add initial radix cache support (#18724)

2026-03-05 15:24:09 +08:00
parent c1df359b44
commit 727face6c2
6 changed files with 83 additions and 56 deletions
--- a/test/registered/dllm/test_llada2_mini.py
+++ b/test/registered/dllm/test_llada2_mini.py
@@ -73,7 +73,7 @@ class TestLLaDA2Mini(CustomTestCase):
        if is_in_amd_ci():
            self.assertGreater(metrics["output_throughput"], 80)
        else:
-            self.assertGreater(metrics["output_throughput"], 250)
+            self.assertGreater(metrics["output_throughput"], 350)

    def test_bs_1_speed(self):
        args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)