Support setting use_thread in the run_program for easier debugging. (#1823)

Co-authored-by: Byron Hsu <byronhsu1230@gmail.com>

python/sglang/lang/interpreter.py

@@ -54,7 +54,14 @@ def run_internal(state, program, func_args, func_kwargs, sync):
 
 
 def run_program(
-    program, backend, func_args, func_kwargs, default_sampling_para, stream, sync=False
+    program,
+    backend,
+    func_args,
+    func_kwargs,
+    default_sampling_para,
+    stream,
+    sync=False,
+    use_thread=True,
 ):
     if hasattr(backend, "endpoint"):
         backend = backend.endpoint
@@ -67,6 +74,7 @@ def run_program(
         chat_template=None,
         stream=stream,
         num_api_spec_tokens=program.num_api_spec_tokens,
+        use_thread=use_thread,
     )
     state = ProgramState(stream_executor)
 

python/sglang/lang/ir.py

@@ -168,6 +168,7 @@ class SglFunction:
         return_text_in_logprobs: Optional[bool] = None,
         stream: bool = False,
         backend=None,
+        use_thread: bool = True,
         **kwargs,
     ):
         from sglang.lang.interpreter import run_program
@@ -195,7 +196,15 @@ class SglFunction:
             return_text_in_logprobs=return_text_in_logprobs,
         )
         backend = backend or global_config.default_backend
-        return run_program(self, backend, args, kwargs, default_sampling_para, stream)
+        return run_program(
+            self,
+            backend,
+            args,
+            kwargs,
+            default_sampling_para,
+            stream,
+            use_thread=use_thread,
+        )
 
     def run_batch(
         self,