[DLLM] Add CI for diffusion LLMs (#14723)

python/sglang/srt/managers/schedule_batch.py

@@ -723,6 +723,7 @@ class Req:
         self.dimensions = dimensions
 
         # For diffusion LLM
+        self.dllm_ids = []
         self.dllm_block_offset = 0
         self.dllm_config = dllm_config
 
@@ -799,14 +800,15 @@ class Req:
         return self.dllm_config is not None
 
     def _init_fill_ids_for_dllm(self):
-        if not self.fill_ids:
-            self.fill_ids = (
+        if not self.dllm_ids:
+            self.dllm_ids = (
                 self.origin_input_ids
                 + [self.dllm_config.mask_id] * self.dllm_config.block_size
             )
         else:
             self.dllm_block_offset += self.dllm_config.block_size
-            self.fill_ids += [self.dllm_config.mask_id] * self.dllm_config.block_size
+            self.dllm_ids += [self.dllm_config.mask_id] * self.dllm_config.block_size
+        self.fill_ids = self.dllm_ids
 
     def init_next_round_input(self, tree_cache: Optional[BasePrefixCache] = None):
         if self.is_dllm():

test/srt/dllm/test_llada2_mini.py

@@ -0,0 +1,81 @@
+import unittest
+from types import SimpleNamespace
+
+from sglang.srt.utils import kill_process_tree
+from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
+from sglang.test.send_one import BenchArgs, send_one_prompt
+from sglang.test.test_utils import (
+    DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+    DEFAULT_URL_FOR_TEST,
+    CustomTestCase,
+    is_in_amd_ci,
+    is_in_ci,
+    popen_launch_server,
+    write_github_step_summary,
+)
+
+
+class TestLLaDA2Mini(CustomTestCase):
+    @classmethod
+    def setUpClass(cls):
+        cls.model = "inclusionAI/LLaDA2.0-mini"
+        cls.base_url = DEFAULT_URL_FOR_TEST
+
+        other_args = [
+            "--trust-remote-code",
+            "--mem-fraction-static",
+            "0.9",
+            "--max-running-requests",
+            "1",
+            "--attention-backend",
+            "flashinfer",
+            "--dllm-algorithm",
+            "LowConfidence",  # TODO: Add dLLM configurations
+        ]
+
+        cls.process = popen_launch_server(
+            cls.model,
+            cls.base_url,
+            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+            other_args=other_args,
+        )
+
+    @classmethod
+    def tearDownClass(cls):
+        kill_process_tree(cls.process.pid)
+
+    def test_gsm8k(self):
+        args = SimpleNamespace(
+            num_shots=5,
+            data_path=None,
+            num_questions=200,
+            max_new_tokens=512,
+            parallel=128,
+            host="http://127.0.0.1",
+            port=int(self.base_url.split(":")[-1]),
+        )
+        metrics = run_eval_few_shot_gsm8k(args)
+        print(f"{metrics=}")
+
+        self.assertGreater(metrics["accuracy"], 0.6)
+        self.assertGreater(metrics["output_throughput"], 150)
+
+    def test_bs_1_speed(self):
+        args = BenchArgs(port=int(self.base_url.split(":")[-1]), max_new_tokens=2048)
+        acc_length, speed = send_one_prompt(args)
+
+        print(f"{speed=:.2f}")
+
+        if is_in_ci():
+            write_github_step_summary(
+                f"### test_bs_1_speed (llada2-mini) with tp1\n"
+                f"{speed=:.2f} token/s\n"
+            )
+            if is_in_amd_ci():
+                self.assertGreater(speed, 10)
+            else:
+                self.assertGreater(speed, 250)
+
+
+if __name__ == "__main__":
+    unittest.main()

test/srt/run_suite.py

@@ -51,6 +51,7 @@ suites = {
         TestFile("rl/test_fp32_lm_head.py", 9),
         # TestFile("rl/test_update_weights_from_disk.py", 210),  # Temporarily disabled, see https://github.com/sgl-project/sglang/pull/13998
         TestFile("rl/test_update_weights_from_tensor.py", 195),
+        TestFile("dllm/test_llada2_mini.py", 520),
         TestFile("test_abort.py", 131),
         TestFile("test_chunked_prefill.py", 312),
         TestFile("test_create_kvindices.py", 7),