Fix dockerfile and triton cache manager (#720)

python/sglang/srt/server.py

@@ -52,6 +52,7 @@ from sglang.srt.utils import (
     allocate_init_ports,
     assert_pkg_version,
     enable_show_time_cost,
+    maybe_set_triton_cache_manager,
     set_ulimit,
 )
 from sglang.utils import get_exception_traceback
@@ -201,6 +202,11 @@ def launch_server(
             "reinstall the latest version by following the instructions "
             "at https://docs.flashinfer.ai/installation.html.",
         )
+
+    if server_args.tp_size // server_args.dp_size > 1:
+        # FIXME: remove this after https://github.com/triton-lang/triton/pull/4295 is used as a dependency.
+        maybe_set_triton_cache_manager()
+
     if server_args.chat_template:
         # TODO: replace this with huggingface transformers template
         load_chat_template_for_openai_api(server_args.chat_template)