Simplify the BatchMultimodalOutput in io_struct.py (#12993)

python/sglang/srt/managers/io_struct.py

@@ -924,7 +924,7 @@ class BatchTokenIDOutput(
 
 
 @dataclass
-class BatchMultimodalDecodeReq(BaseBatchReq, RequestTimingMetricsMixin):
+class BatchMultimodalDecodeReq(BaseBatchReq):
     decoded_ids: List[int]
     input_token_logprobs_val: List[float]
     input_token_logprobs_idx: List[int]
@@ -1003,7 +1003,7 @@ class BatchStrOutput(
 
 
 @dataclass
-class BatchMultimodalOutput(BaseBatchReq, RequestTimingMetricsMixin):
+class BatchMultimodalOutput(BaseBatchReq):
     # The finish reason
     finished_reasons: List[dict]
     decoded_ids: List[List[int]]

python/sglang/srt/managers/scheduler_output_processor_mixin.py

@@ -940,6 +940,8 @@ class SchedulerOutputProcessorMixin:
 
             self.send_to_detokenizer.send_output(
                 BatchTokenIDOutput(
+                    rids=rids,
+                    http_worker_ipcs=http_worker_ipcs,
                     spec_verify_ct=spec_verify_ct,
                     spec_accepted_tokens=spec_accepted_tokens,
                     queue_time=queue_times,
@@ -971,8 +973,6 @@ class SchedulerOutputProcessorMixin:
                     output_token_ids_logprobs_idx=output_token_ids_logprobs_idx,
                     output_token_entropy_val=None,
                     output_hidden_states=output_hidden_states,
-                    rids=rids,
-                    http_worker_ipcs=http_worker_ipcs,
                     placeholder_tokens_idx=None,
                     placeholder_tokens_val=None,
                     retraction_counts=retraction_counts,
@@ -1025,6 +1025,8 @@ class SchedulerOutputProcessorMixin:
                 retraction_counts.append(req.retraction_count)
         self.send_to_detokenizer.send_output(
             BatchEmbeddingOutput(
+                rids=rids,
+                http_worker_ipcs=http_worker_ipcs,
                 queue_time=queue_times,
                 forward_entry_time=forward_entry_times,
                 prefill_delay=prefill_delays,
@@ -1033,10 +1035,8 @@ class SchedulerOutputProcessorMixin:
                 embeddings=embeddings,
                 prompt_tokens=prompt_tokens,
                 cached_tokens=cached_tokens,
-                http_worker_ipcs=http_worker_ipcs,
                 placeholder_tokens_idx=None,
                 placeholder_tokens_val=None,
                 retraction_counts=retraction_counts,
-                rids=rids,
             )
         )

test/srt/test_request_queue_validation.py

@@ -65,9 +65,10 @@ class TestMaxQueuedRequests(CustomTestCase):
         status_codes = asyncio.run(
             send_concurrent_generate_requests(self.base_url, num_requests=10)
         )
+        self.assertLessEqual(status_codes.count(200), 2)
 
-        expected_status_codes = [200, 200, 503, 503, 503, 503, 503, 503, 503, 503]
+        # expected_status_codes = [200, 200, 503, 503, 503, 503, 503, 503, 503, 503]
-        assert status_codes == expected_status_codes
+        # self.assertEqual(status_codes, expected_status_codes)
 
     def test_max_running_requests_and_max_queued_request_validation(self):
         """Verify running request and queued request numbers based on server logs."""