[ET-VK] Enable Partial GPU lowering via Vulkan in stories model export

Pull Request resolved: #2368

## Context

Simple change to add Vulkan Partitioner as a dependency for the llama exporter and runner, and provide a command line flag to invoke the vulkan partitioner during export.

ghstack-source-id: 218421177
@exported-using-ghexport

Differential Revision: [D54805831](https://our.internmc.facebook.com/intern/diff/D54805831/)

Author: SS-JIA

examples/models/llama2/TARGETS

@@ -82,6 +82,7 @@ runtime.python_library(
         "//executorch/backends/transforms:duplicate_dynamic_quant_chain",
         "//executorch/backends/xnnpack:xnnpack_backend",
         "//executorch/backends/xnnpack/partition:xnnpack_partitioner",
+        "//executorch/backends/vulkan/partitioner:vulkan_partitioner",
         "//executorch/examples/models:model_base",
         "//executorch/examples/models:models",
         "//executorch/examples/portable:utils",

examples/models/llama2/export_llama_lib.py

@@ -17,6 +17,7 @@
 
 import pkg_resources
 import torch
+from executorch.backends.vulkan.partitioner.vulkan_partitioner import VulkanPartitioner
 from executorch.backends.xnnpack.partition.xnnpack_partitioner import (
     XnnpackDynamicallyQuantizedPartitioner,
 )
@@ -356,6 +357,7 @@ def build_args_parser() -> argparse.ArgumentParser:
     parser.add_argument("-2", "--fairseq2", action="store_true")
     parser.add_argument("-v", "--verbose", action="store_true")
     parser.add_argument("-X", "--xnnpack", action="store_true")
+    parser.add_argument("-V", "--vulkan", action="store_true")
 
     return parser
 
@@ -451,6 +453,9 @@ def _export_llama(modelname, args) -> str:  # noqa: C901
         )
         # partitioners[XnnpackPartitioner.__name__] = XnnpackPartitioner()
         modelname = f"xnnpack_{modelname}"
+    if args.vulkan:
+        partitioners[VulkanPartitioner.__name__] = VulkanPartitioner()
+        modelname = f"vulkan_{modelname}"
 
     builder = (
         load_llama_model(

examples/models/llama2/runner/targets.bzl

@@ -36,7 +36,11 @@ def define_common_targets():
                 "//executorch/extension/module:module" + aten_suffix,
                 "//executorch/kernels/quantized:generated_lib" + aten_suffix,
                 "//executorch/runtime/core/exec_aten:lib" + aten_suffix,
-            ] + (_get_operator_lib(aten)),
+            ] + (_get_operator_lib(aten)) + ([
+                # Vulkan API currently cannot build on some platforms (e.g. Apple, FBCODE)
+                # Therefore enable it explicitly for now to avoid failing tests
+                "//executorch/backends/vulkan:vulkan_backend_lib",
+            ] if native.read_config("llama", "use_vulkan", "0") == "1" else []),
             external_deps = [
                 "libtorch",
             ] if aten else [],