[Feature, Hardware] Enable SGLang on AMD GPUs via PyTorch for ROCm (#1420)

python/sglang/srt/server.py

@@ -78,6 +78,7 @@ from sglang.srt.utils import (
     assert_pkg_version,
     configure_logger,
     enable_show_time_cost,
+    is_hip,
     kill_child_process,
     maybe_set_triton_cache_manager,
     prepare_model,
@@ -434,6 +435,10 @@ def _set_envs_and_config(server_args: ServerArgs):
             "at https://docs.flashinfer.ai/installation.html.",
         )
 
+    if is_hip():
+        # to figure out a better method of not using fork later
+        mp.set_start_method("spawn", force=True)
+
 
 def _wait_and_warmup(server_args, pipe_finish_writer, pid):
     headers = {}