diff --git a/extension/llm/batching/BUCK b/extension/llm/batching/BUCK
new file mode 100644
index 00000000000..1e8cc179228
--- /dev/null
+++ b/extension/llm/batching/BUCK
@@ -0,0 +1,8 @@
+# Any targets that should be shared between fbcode and xplat must be defined in
+# targets.bzl. This file can contain xplat-only targets.
+
+load(":targets.bzl", "define_common_targets")
+
+oncall("executorch")
+
+define_common_targets()
diff --git a/extension/llm/batching/targets.bzl b/extension/llm/batching/targets.bzl
new file mode 100644
index 00000000000..ec5c6ad0818
--- /dev/null
+++ b/extension/llm/batching/targets.bzl
@@ -0,0 +1,54 @@
+load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
+
+def define_common_targets():
+    """Mirrors the extension_llm_batching CMake targets.
+
+    `batching` is the scheduler, the executor seam and the runner, free of
+    ExecuTorch runtime types. `module_executor` implements the seam against a
+    program and a KV cache.
+    """
+    runtime.cxx_library(
+        name = "batching",
+        srcs = [
+            "metrics.cpp",
+            "runner.cpp",
+        ],
+        exported_headers = [
+            "decode_first_scheduler.h",
+            "executor.h",
+            "metrics.h",
+            "prefix_cache.h",
+            "runner.h",
+            "scheduler.h",
+            "types.h",
+        ],
+        visibility = ["PUBLIC"],
+        exported_deps = [
+            "//executorch/runtime/platform:compiler",
+        ],
+    )
+
+    runtime.cxx_library(
+        name = "module_executor",
+        srcs = [
+            "module_executor.cpp",
+        ],
+        exported_headers = [
+            "module_executor.h",
+        ],
+        visibility = ["PUBLIC"],
+        exported_deps = [
+            ":batching",
+            "//executorch/extension/llm/cache:kv_cache",
+            "//executorch/extension/llm/runner:stats",
+            "//executorch/extension/module:module",
+            "//executorch/runtime/core:core",
+        ],
+        deps = [
+            "//executorch/extension/llm/sampler:sampler",
+            "//executorch/extension/tensor:tensor",
+            "//executorch/runtime/backend:interface",
+            "//executorch/runtime/core/exec_aten/util:scalar_type_util",
+            "//executorch/runtime/platform:platform",
+        ],
+    )
diff --git a/extension/llm/batching/test/BUCK b/extension/llm/batching/test/BUCK
new file mode 100644
index 00000000000..1e8cc179228
--- /dev/null
+++ b/extension/llm/batching/test/BUCK
@@ -0,0 +1,8 @@
+# Any targets that should be shared between fbcode and xplat must be defined in
+# targets.bzl. This file can contain xplat-only targets.
+
+load(":targets.bzl", "define_common_targets")
+
+oncall("executorch")
+
+define_common_targets()
diff --git a/extension/llm/batching/test/targets.bzl b/extension/llm/batching/test/targets.bzl
new file mode 100644
index 00000000000..fbab7d5326d
--- /dev/null
+++ b/extension/llm/batching/test/targets.bzl
@@ -0,0 +1,18 @@
+load("@fbsource//xplat/executorch/build:runtime_wrapper.bzl", "runtime")
+
+def define_common_targets():
+    """Mirrors extension_llm_batching_test in CMakeLists.txt."""
+    runtime.cxx_test(
+        name = "test",
+        srcs = [
+            "prefix_cache_test.cpp",
+            "runner_test.cpp",
+            "scheduler_test.cpp",
+        ],
+        headers = [
+            "fake_executor.h",
+        ],
+        deps = [
+            "//executorch/extension/llm/batching:batching",
+        ],
+    )