Improve the coverage of the openai api server test (#878)

python/sglang/srt/layers/logits_processor.py

@@ -209,7 +209,7 @@ class LogitsProcessor(nn.Module):
                 all_logits = all_logits[:, : self.config.vocab_size].float()
 
                 all_logprobs = all_logits
-                del all_logits
+                del all_logits, hidden_states
                 all_logprobs[:] = torch.nn.functional.log_softmax(all_logprobs, dim=-1)
 
                 # Get the logprob of top-k tokens

python/sglang/srt/server.py

@@ -72,8 +72,8 @@ from sglang.srt.utils import (
     allocate_init_ports,
     assert_pkg_version,
     enable_show_time_cost,
-    maybe_set_triton_cache_manager,
     kill_child_process,
+    maybe_set_triton_cache_manager,
     set_ulimit,
 )
 from sglang.utils import get_exception_traceback