diff --git a/.bazelrc b/.bazelrc
index 6278d956d75..c41bf62af6f 100644
--- a/.bazelrc
+++ b/.bazelrc
@@ -17,7 +17,8 @@ build:cuda_clang --host_copt=-Wno-error=unused-command-line-argument
 build:cuda_clang --repo_env TF_NEED_TENSORRT=0
 build:cuda_clang --action_env=TF_CUDA_CLANG="1"
 build:cuda_clang --@local_config_cuda//:cuda_compiler=clang
-build:cuda_clang --repo_env=TF_CUDA_COMPUTE_CAPABILITIES="sm_60,sm_70,sm_80,sm_89,compute_90"
+build:cuda_clang --repo_env=TF_CUDA_COMPUTE_CAPABILITIES="sm_60,sm_70,sm_75,sm_80,sm_86,sm_89,compute_90"
+build:cuda_clang --copt=-Dtypeof=__typeof__
 
 common:cuda_clang --repo_env=HERMETIC_CUDA_VERSION="12.5.1"
 common:cuda_clang --repo_env=HERMETIC_CUDNN_VERSION="9.3.0"
@@ -116,6 +117,9 @@ common --repo_env USE_HERMETIC_CC_TOOLCHAIN=1
 common --repo_env=HERMETIC_PYTHON_VERSION="3.10"
 common --repo_env=USE_PYWRAP_RULES=1
 build --features=-layering_check
+build --host_features=-layering_check
+build --features=-module_maps
+build --host_features=-module_maps
 build --define=framework_shared_object=false
 build --dynamic_mode=off
 
diff --git a/WORKSPACE b/WORKSPACE
index 9d53b2fbf8b..ddbd10ac0eb 100644
--- a/WORKSPACE
+++ b/WORKSPACE
@@ -24,8 +24,8 @@ local_repository(
 load("//tensorflow_serving:repo.bzl", "tensorflow_http_archive")
 tensorflow_http_archive(
     name = "org_tensorflow",
-    sha256 = "54bfd56d61dfced2b82bdeee3262a24fb3d59bed4dbb292d19834af002d85b3b",
-    git_commit = "c1be979f3130b85f14e8890c05cbca147784f474",
+    sha256 = "6438396f3b19af5d7ad787cf041f857af7505916dc08092e20b07d1b1f8df492",
+    git_commit = "a481b10260dfdf833a1b16007eead49c1d7febf3",
     patch = "//third_party/tensorflow:tensorflow.patch",
     patch_cmds = [
         "sed -i '/cc_library = _cc_library/d' tensorflow/core/platform/rules_cc.bzl",
@@ -95,17 +95,16 @@ http_archive(
 
 http_archive(
     name = "rules_python",
-    sha256 = "8964aa1e7525fea5244ba737458694a057ada1be96a92998a41caa1983562d00",
-    strip_prefix = "rules_python-1.8.5",
+    sha256 = "fa7dd2c6b7d63b3585028dd8a90a6cf9db83c33b250959c2ee7b583a6c130e12",
+    strip_prefix = "rules_python-1.6.0",
     urls = [
-        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/bazelbuild/rules_python/releases/download/1.8.5/rules_python-1.8.5.tar.gz",
-        "https://github.com/bazelbuild/rules_python/releases/download/1.8.5/rules_python-1.8.5.tar.gz",
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/bazelbuild/rules_python/releases/download/1.6.0/rules_python-1.6.0.tar.gz",
+        "https://github.com/bazelbuild/rules_python/releases/download/1.6.0/rules_python-1.6.0.tar.gz",
     ],
     patches = [
-        "@rules_ml_toolchain//third_party/rules_python:rules_python_scope.patch",
-        "@rules_ml_toolchain//third_party/rules_python:rules_python_freethreaded.patch",
-        "@rules_ml_toolchain//third_party/rules_python:rules_python_versions.patch",
-        "@rules_ml_toolchain//third_party/rules_python:rules_python_pip_version.patch",
+        "@org_tensorflow//third_party/py:rules_python_pip_version.patch",
+        "@org_tensorflow//third_party/py:rules_python_freethreaded.patch",
+        "@org_tensorflow//third_party/py:rules_python_versions.patch",
     ],
     patch_args = ["-p1"],
 )
@@ -114,11 +113,15 @@ http_archive(
 # Details: https://github.com/google-ml-infra/rules_ml_toolchain
 http_archive(
     name = "rules_ml_toolchain",
-    sha256 = "0b42f693a60c6050d87db1e0a0eaeb84ab3f54191fce094d86334faedc807da0",
-    strip_prefix = "rules_ml_toolchain-398d613aea7a4c294da49b79a6d6f3f8732bd84c",
+    sha256 = "54c1a357f71f611efdb4891ebd4bcbe4aeb6dfa7e473f14fd7ecad5062096616",
+    strip_prefix = "rules_ml_toolchain-d8cb9c2c168cd64000eaa6eda0781a9615a26ffe",
     urls = [
-        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/google-ml-infra/rules_ml_toolchain/archive/398d613aea7a4c294da49b79a6d6f3f8732bd84c.tar.gz",
-        "https://github.com/google-ml-infra/rules_ml_toolchain/archive/398d613aea7a4c294da49b79a6d6f3f8732bd84c.tar.gz",
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/google-ml-infra/rules_ml_toolchain/archive/d8cb9c2c168cd64000eaa6eda0781a9615a26ffe.tar.gz",
+        "https://github.com/google-ml-infra/rules_ml_toolchain/archive/d8cb9c2c168cd64000eaa6eda0781a9615a26ffe.tar.gz",
+    ],
+    patch_cmds = [
+        "mkdir -p third_party/rules_python py && touch third_party/rules_python/BUILD py/BUILD",
+        "cat << 'EOF' >> cc/cuda/clang/clang_cuda_runtime_wrapper.h\n#if defined(__clang__) && defined(__CUDA__)\n#include <vector_types.h>\n#ifndef typeof\n#define typeof __typeof__\n#endif\n#ifndef DISABLE_WRAPPERS_STWT\n__device__ inline void __stwt(uint4* ptr, uint4 value) {\n  asm volatile(\"st.global.wt.v4.u32 [%0], {%1, %2, %3, %4};\" : : \"l\"(ptr), \"r\"(value.x), \"r\"(value.y), \"r\"(value.z), \"r\"(value.w) : \"memory\");\n}\n#endif\n#endif\nEOF",
     ],
 )
 
@@ -146,7 +149,6 @@ python_init_repositories(
         "3.11": "@org_tensorflow//:requirements_lock_3_11.txt",
         "3.12": "@org_tensorflow//:requirements_lock_3_12.txt",
         "3.13": "@org_tensorflow//:requirements_lock_3_13.txt",
-        "3.14": "@org_tensorflow//:requirements_lock_3_14.txt",
     },
 )
 
diff --git a/tensorflow_serving/opensource_only/.bazelrc b/tensorflow_serving/opensource_only/.bazelrc
new file mode 100644
index 00000000000..c41bf62af6f
--- /dev/null
+++ b/tensorflow_serving/opensource_only/.bazelrc
@@ -0,0 +1,131 @@
+# Optimizations used for TF Serving release builds.
+build:release --copt=-mavx
+build:release --copt=-msse4.2
+
+# Options used to build with CUDA.
+build:cuda --repo_env TF_NEED_CUDA=1
+build:cuda --crosstool_top=@local_config_cuda//crosstool:toolchain
+build:cuda --@local_config_cuda//:enable_cuda
+
+# Options used to build with CUDA clang
+build:cuda_clang --config=cuda
+build:cuda_clang --copt=-Wno-gnu-offsetof-extensions
+build:cuda_clang --copt=-Wno-error=unused-command-line-argument
+build --linkopt=-Wno-unused-command-line-argument
+build --host_linkopt=-Wno-unused-command-line-argument
+build:cuda_clang --host_copt=-Wno-error=unused-command-line-argument
+build:cuda_clang --repo_env TF_NEED_TENSORRT=0
+build:cuda_clang --action_env=TF_CUDA_CLANG="1"
+build:cuda_clang --@local_config_cuda//:cuda_compiler=clang
+build:cuda_clang --repo_env=TF_CUDA_COMPUTE_CAPABILITIES="sm_60,sm_70,sm_75,sm_80,sm_86,sm_89,compute_90"
+build:cuda_clang --copt=-Dtypeof=__typeof__
+
+common:cuda_clang --repo_env=HERMETIC_CUDA_VERSION="12.5.1"
+common:cuda_clang --repo_env=HERMETIC_CUDNN_VERSION="9.3.0"
+common:cuda_clang --repo_env=HERMETIC_NVSHMEM_VERSION="3.2.5"
+common:cuda_clang --repo_env=HERMETIC_NCCL_VERSION="2.29.7"
+build:cuda_clang --action_env=GCC_HOST_COMPILER_PATH="/usr/bin/gcc-10"
+build:cuda_clang --action_env=CLANG_CUDA_COMPILER_PATH="/usr/lib/llvm-17/bin/clang"
+
+# Options used to build with TPU support.
+build:tpu --define=with_tpu_support=true --define=framework_shared_object=false
+build:tpu --copt=-DLIBTPU_ON_GCE
+
+# Please note that MKL on MacOS or windows is still not supported.
+# If you would like to use a local MKL instead of downloading, please set the
+# environment variable "TF_MKL_ROOT" every time before build.
+build:mkl --define=build_with_mkl=true --define=enable_mkl=true --define=build_with_openmp=false
+build:mkl --define=tensorflow_mkldnn_contraction_kernel=0
+
+# This config option is used to enable MKL-DNN open source library only,
+# without depending on MKL binary version.
+build:mkl_open_source_only --define=build_with_mkl_dnn_only=true
+build:mkl_open_source_only --define=build_with_mkl=true --define=enable_mkl=true
+build:mkl_open_source_only --define=tensorflow_mkldnn_contraction_kernel=0
+
+# Config setting to build oneDNN with Compute Library for the Arm Architecture (ACL).
+# This build is for the inference regime only.
+build:mkl_aarch64 --define=build_with_mkl_aarch64=true --define=enable_mkl=true
+build:mkl_aarch64 --define=tensorflow_mkldnn_contraction_kernel=0
+build:mkl_aarch64 --define=build_with_mkl_opensource=true
+build:mkl_aarch64 --define=build_with_openmp=true
+build:mkl_aarch64 --copt=-march=armv8.2-a
+build:mkl_aarch64 --copt=-O3
+
+build --define=build_with_onednn_v2=true
+build --define=xnn_enable_avxvnni=false
+build --define=xnn_enable_avxvnniint8=false
+build --define=xnn_enable_avx256vnnigfni=false
+build --define=xnn_enable_avx512amx=false
+build --define=xnn_enable_avx512fp16=false
+
+# Processor native optimizations (depends on build host capabilities).
+build:nativeopt --copt=-march=native
+build:nativeopt --host_copt=-march=native
+build:nativeopt --copt=-O3
+
+build --keep_going
+build --verbose_failures=true
+build --spawn_strategy=standalone
+build --genrule_strategy=standalone
+
+build --define=grpc_no_ares=true
+
+# Sets the default Apple platform to macOS.
+build --apple_platform_type=macos
+
+build -c opt
+
+# LLVM, MLIR and TF requires C++17.
+build --cxxopt=-std=c++17
+build --host_cxxopt=-std=c++17
+
+# Adding "--cxxopt=-D_GLIBCXX_USE_CXX11_ABI=0" creates parity with TF
+# compilation options. It also addresses memory use due to
+# copy-on-write semantics of std::strings of the older ABI.
+build --cxxopt=-D_GLIBCXX_USE_CXX11_ABI=0
+
+build --workspace_status_command=/proc/self/cwd/tools/gen_status_stamp.sh
+
+build --experimental_repo_remote_exec
+
+# TF now has `cc_shared_library` targets, so it needs the experimental flag
+build --experimental_cc_shared_library
+
+# Yggdrasil Decision Forests (internal library of TensorFlow Decision Forests)
+# uses TensorFlow for all IO operations.
+build --define=use_tensorflow_io=1
+
+# TensorFlow Decision Forests does not use Absl concurrency primitives on MacOs.
+# Reason: TensorFlow/ABSL ODR violation (b/214189609) # copybara:strip
+build:macos --define std_synchronization_primitives=1
+
+# Taken from https://github.com/openxla/xla/blob/99559d7a4f7c55490f46385ad29a3cbf9c3911af/warnings.bazelrc#L6
+# We silence warnings for code in `external`.
+build --per_file_copt=external/.*@-w
+build --host_per_file_copt=external/.*@-w
+build --copt=-Wno-macro-redefined # absl vs tsl logging clash
+build --copt=-Wno-sign-compare # int as loop variable
+build --copt=-Wno-deprecated-declarations
+build --copt=-Wno-unused-but-set-variable # due to `ifdefs` in ml_dtypes
+
+# Revert to the legacy WORKSPACE dependency resolution system
+# and disable Bzlmod.
+common --enable_bzlmod=false
+common --incompatible_enable_cc_toolchain_resolution
+common --repo_env USE_HERMETIC_CC_TOOLCHAIN=1
+common --repo_env=HERMETIC_PYTHON_VERSION="3.10"
+common --repo_env=USE_PYWRAP_RULES=1
+build --features=-layering_check
+build --host_features=-layering_check
+build --features=-module_maps
+build --host_features=-module_maps
+build --define=framework_shared_object=false
+build --dynamic_mode=off
+
+
+common:clang_local --noincompatible_enable_cc_toolchain_resolution
+common:clang_local --noincompatible_enable_android_toolchain_resolution
+common:clang_local --@rules_ml_toolchain//cc_toolchain/config:enable_hermetic_cc=False
+common:clang_local --repo_env USE_HERMETIC_CC_TOOLCHAIN=0
+
diff --git a/tensorflow_serving/opensource_only/root.workspace b/tensorflow_serving/opensource_only/root.workspace
new file mode 100644
index 00000000000..ddbd10ac0eb
--- /dev/null
+++ b/tensorflow_serving/opensource_only/root.workspace
@@ -0,0 +1,296 @@
+workspace(name = "tf_serving")
+
+# TODO(b/269515133): We temporarily remove remote_predict from our builds for
+# 2.12 due to a breakage caused by
+# github.com/tensorflow/tensorflow/commit/6147c03eb9af1e5d2ae155045b33e909ef96944e
+# This will be removed in a subsequent release.
+local_repository(
+    name = "ignore_remote_predict",
+    path = "tensorflow_serving/experimental/tensorflow/ops/remote_predict/",
+)
+
+# ===== TensorFlow dependency =====
+#
+# TensorFlow is imported here instead of in tf_serving_workspace() because
+# existing automation scripts that bump the TF commit hash expect it here.
+#
+# To update TensorFlow to a new revision.
+# 1. Update the 'git_commit' args below to include the new git hash.
+# 2. Get the sha256 hash of the archive with a command such as...
+#    curl -L https://github.com/tensorflow/tensorflow/archive/<git hash>.tar.gz | sha256sum
+#    and update the 'sha256' arg with the result.
+# 3. Request the new archive to be mirrored on mirror.bazel.build for more
+#    reliable downloads.
+load("//tensorflow_serving:repo.bzl", "tensorflow_http_archive")
+tensorflow_http_archive(
+    name = "org_tensorflow",
+    sha256 = "6438396f3b19af5d7ad787cf041f857af7505916dc08092e20b07d1b1f8df492",
+    git_commit = "a481b10260dfdf833a1b16007eead49c1d7febf3",
+    patch = "//third_party/tensorflow:tensorflow.patch",
+    patch_cmds = [
+        "sed -i '/cc_library = _cc_library/d' tensorflow/core/platform/rules_cc.bzl",
+        "echo -e \"\\ndef cc_library_oss(deps=[], **kwargs):\\n    if kwargs.get(\\\"name\\\") == \\\"lib_internal_impl\\\" or \\\"protobuf\\\" in kwargs.get(\\\"name\\\", \\\"\\\"):\\n        _cc_library(deps = deps, **kwargs)\\n        return\\n    if type(deps) == \\\"list\\\":\\n        if \\\"@com_google_protobuf//:protobuf\\\" not in deps:\\n            deps = deps + [\\\"@com_google_protobuf//:protobuf\\\"]\\n    else:\\n        deps = deps + [\\\"@com_google_protobuf//:protobuf\\\"]\\n    _cc_library(deps = deps, **kwargs)\\ncc_library = cc_library_oss\" >> tensorflow/core/platform/rules_cc.bzl",
+        "sed -i 's#deps = \\[op_gen\\] + deps#deps = [op_gen] + deps + [clean_dep(\"//tensorflow/core/framework:kernel_shape_util\"), clean_dep(\"//tensorflow/core/framework:full_type_util\")]#' tensorflow/tensorflow.bzl",
+        "sed -i '/name = \"kernel_shape_util\",/a \\    visibility = [\"//visibility:public\"],' tensorflow/core/framework/BUILD",
+        "echo -e '\\nalias(name = \"tensorflow_libtensorflow_framework\", actual = \"//tensorflow/core:tensorflow\", visibility = [\"//visibility:public\"])' >> BUILD",
+        "echo -e '\\nalias(name = \"tensorflow_tf_header_lib\", actual = \"//tensorflow/core:tensorflow\", visibility = [\"//visibility:public\"])' >> BUILD",
+        "sed -i '/name = \"env\",/,/deps = \\[/ s#deps = \\[#deps = [\":status\", \":statusor\", \":context\", \":tracing\", \"//xla/tsl/profiler/backends/cpu:threadpool_listener_state\", \"//xla/tsl/platform:byte_order\", #' third_party/xla/xla/tsl/platform/default/BUILD",
+        "sed -i '/name = \"tracing\",/,/deps = \\[/ s#deps = \\[#deps = [\"//xla/tsl/platform:logging\", #' third_party/xla/xla/tsl/platform/default/BUILD",
+        "sed -i '/name = \"tf_runtime\",/a \\        repo_mapping = {\"@xla\": \"@local_xla\", \"@tsl\": \"@local_tsl\"},' third_party/tf_runtime/workspace.bzl",
+        "sed -i '/name = \"error_util\",/,/deps = \\[/ s#deps = \\[#deps = [\"@xla//xla/tsl/concurrency:async_value\", \"@xla//xla/tsl/concurrency:concurrent_vector\", \"@xla//xla/tsl/concurrency:executor\", \"@xla//xla/tsl/concurrency:ref_count\", \"@xla//xla/tsl/util:safe_reinterpret_cast\", \"@tsl//tsl/platform:context\", #' tensorflow/core/tfrt/utils/BUILD",
+        "sed -i '/name = \"work_queue_interface\",/,/deps = \\[/ s#deps = \\[#deps = [\"@xla//xla/tsl/concurrency:ref_count\", #' tensorflow/core/tfrt/runtime/BUILD",
+        "sed -i '/name = \"execute\",/,/deps = \\[/ s#deps = \\[#deps = [\"@xla//xla/tsl/platform:macros\", \"@xla//xla/tsl/platform:types\", \"@xla//xla/tsl/profiler/utils:no_init\", \"@tsl//tsl/profiler/lib:traceme_encode\", \"@xla//xla/tsl/profiler/utils:traceme_global_flags\", \"@xla//xla/tsl/profiler/backends/cpu:traceme_recorder\", \"@tsl//tsl/platform:bfloat16\", \"@tsl//tsl/platform:ml_dtypes\", \"@tsl//tsl/platform:tstring\", \"@tsl//tsl/platform:cord\", \"@tsl//tsl/platform:refcount\", \"@tsl//tsl/platform:thread_annotations\", \"@tsl//tsl/platform:stringpiece\", \"@xla//xla/tsl/profiler/utils:time_utils\", \"@xla//xla/tsl/profiler/utils:math_utils\", #' tensorflow/core/tfrt/mlrt/interpreter/BUILD",
+        "sed -i '/tf_vendored(name = \"xla\",/s/)/, repo_mapping = {\"@xla\": \"@local_xla\", \"@tsl\": \"@local_tsl\"})/' tensorflow/workspace3.bzl",
+        "sed -i '/tf_vendored(name = \"tsl\",/s/)/, repo_mapping = {\"@xla\": \"@local_xla\", \"@tsl\": \"@local_tsl\"})/' tensorflow/workspace3.bzl",
+        """python3 -c 'import re, glob
+for p in glob.glob("third_party/xla/**/BUILD*", recursive=True):
+    s = open(p).read(); blocks = s.split("cc_library(");
+    for i in range(1, len(blocks)):
+        b = blocks[i]; m_th = re.search(r"textual_hdrs\\s*=\\s*(\\[[^\\]]+\\]),?\\n?", b);
+        if m_th:
+            th = m_th.group(1); b = b.replace(m_th.group(0), ""); m_h = re.search(r"hdrs\\s*=\\s*(\\[[^\\]]+\\])", b);
+            if m_h: h = m_h.group(1); merged = h[:-1] + ", " + th[1:]; b = b.replace(m_h.group(0), "hdrs = " + merged);
+            else: b = "\\n    hdrs = " + th + "," + b;
+            blocks[i] = b
+    open(p, "w").write("cc_library(".join(blocks))'""",
+        "echo -e '\\ndiff --git a/WORKSPACE b/WORKSPACE\\n--- a/WORKSPACE\\n+++ b/WORKSPACE\\n@@ -184,25 +184,2 @@\\n sass_repositories()\\n \\n-http_archive(\\n-    name = \"xla\",\\n-    patch_args = [\"-p1\"],\\n-    patches = [\\n-        \"//third_party:xla.patch\",\\n-        \"//third_party:xla_add_grpc_cares_darwin_arm64_support.patch\",\\n-    ],\\n-    sha256 = \"ba80ef58f89ca11bc5652e936cf856cdeae91e6b723ce6750e9ce0202cab51ac\",\\n-    strip_prefix = \"xla-f094066398e2c884e994711fd677f68864324614\",\\n-    urls = [\\n-        \"https://github.com/openxla/xla/archive/f094066398e2c884e994711fd677f68864324614.zip\",\\n-    ],\\n-)\\n-\\n-http_archive(\\n-    name = \"tsl\",\\n-    sha256 = \"8cf1e1285c7b1843a7f5f787465c1ef80304b3400ed837870bc76d74ce04f5af\",\\n-    strip_prefix = \"tsl-d71df2f7612583617d359c36243695097dd63726\",\\n-    urls = [\\n-        \"https://github.com/google/tsl/archive/d71df2f7612583617d359c36243695097dd63726.zip\",\\n-    ],\\n-)\\n-\\n load(\"@xla//tools/toolchains/python:python_repo.bzl\", \"python_repository\")' >> third_party/xprof/xprof.patch",
+    ],
+    repo_mapping = {
+        "@local_xla": "@local_xla",
+        "@local_tsl": "@local_tsl",
+        "@org_tensorflow": "@org_tensorflow",
+        "@xla": "@local_xla",
+        "@tsl": "@local_tsl",
+    },
+)
+
+# Import all of TensorFlow Serving's external dependencies.
+# Downstream projects (projects importing TensorFlow Serving) need to
+# duplicate all code below in their WORKSPACE file in order to also initialize
+# those external dependencies.
+load("//tensorflow_serving:workspace.bzl", "tf_serving_workspace")
+tf_serving_workspace()
+
+# Check bazel version requirement, which is stricter than TensorFlow's.
+load("@bazel_skylib//lib:versions.bzl", "versions")
+versions.check("7.4.1")
+
+load("@bazel_tools//tools/build_defs/repo:http.bzl", "http_archive")
+
+http_archive(
+    name = "bazel_skylib",
+    sha256 = "97e709db2e97b646263b5c5e83e3b00de48c1ae55b9e421e3b5e3f9467d02a3a",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/bazelbuild/bazel-skylib/releases/download/1.8.0/bazel-skylib-1.8.0.tar.gz",
+        "https://github.com/bazelbuild/bazel-skylib/releases/download/1.8.0/bazel-skylib-1.8.0.tar.gz",
+    ],
+)
+
+http_archive(
+    name = "rules_cc",
+    sha256 = "b8b918a85f9144c01f6cfe0f45e4f2838c7413961a8ff23bc0c6cdf8bb07a3b6",
+    strip_prefix = "rules_cc-0.1.5",
+    url = "https://github.com/bazelbuild/rules_cc/releases/download/0.1.5/rules_cc-0.1.5.tar.gz",
+)
+
+http_archive(
+    name = "rules_python",
+    sha256 = "fa7dd2c6b7d63b3585028dd8a90a6cf9db83c33b250959c2ee7b583a6c130e12",
+    strip_prefix = "rules_python-1.6.0",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/bazelbuild/rules_python/releases/download/1.6.0/rules_python-1.6.0.tar.gz",
+        "https://github.com/bazelbuild/rules_python/releases/download/1.6.0/rules_python-1.6.0.tar.gz",
+    ],
+    patches = [
+        "@org_tensorflow//third_party/py:rules_python_pip_version.patch",
+        "@org_tensorflow//third_party/py:rules_python_freethreaded.patch",
+        "@org_tensorflow//third_party/py:rules_python_versions.patch",
+    ],
+    patch_args = ["-p1"],
+)
+
+# Toolchains for ML projects hermetic builds.
+# Details: https://github.com/google-ml-infra/rules_ml_toolchain
+http_archive(
+    name = "rules_ml_toolchain",
+    sha256 = "54c1a357f71f611efdb4891ebd4bcbe4aeb6dfa7e473f14fd7ecad5062096616",
+    strip_prefix = "rules_ml_toolchain-d8cb9c2c168cd64000eaa6eda0781a9615a26ffe",
+    urls = [
+        "https://storage.googleapis.com/mirror.tensorflow.org/github.com/google-ml-infra/rules_ml_toolchain/archive/d8cb9c2c168cd64000eaa6eda0781a9615a26ffe.tar.gz",
+        "https://github.com/google-ml-infra/rules_ml_toolchain/archive/d8cb9c2c168cd64000eaa6eda0781a9615a26ffe.tar.gz",
+    ],
+    patch_cmds = [
+        "mkdir -p third_party/rules_python py && touch third_party/rules_python/BUILD py/BUILD",
+        "cat << 'EOF' >> cc/cuda/clang/clang_cuda_runtime_wrapper.h\n#if defined(__clang__) && defined(__CUDA__)\n#include <vector_types.h>\n#ifndef typeof\n#define typeof __typeof__\n#endif\n#ifndef DISABLE_WRAPPERS_STWT\n__device__ inline void __stwt(uint4* ptr, uint4 value) {\n  asm volatile(\"st.global.wt.v4.u32 [%0], {%1, %2, %3, %4};\" : : \"l\"(ptr), \"r\"(value.x), \"r\"(value.y), \"r\"(value.z), \"r\"(value.w) : \"memory\");\n}\n#endif\n#endif\nEOF",
+    ],
+)
+
+load(
+    "@rules_ml_toolchain//cc/deps:cc_toolchain_deps.bzl",
+    "cc_toolchain_deps",
+)
+
+cc_toolchain_deps()
+
+register_toolchains("@rules_ml_toolchain//cc:linux_x86_64_linux_x86_64")
+register_toolchains("@rules_ml_toolchain//cc:linux_x86_64_linux_x86_64_cuda")
+# register_toolchains("@rules_ml_toolchain//cc:linux_aarch64_linux_aarch64")
+# register_toolchains("@rules_ml_toolchain//cc:linux_aarch64_linux_aarch64_cuda")
+
+# Initialize hermetic Python
+load("@org_tensorflow//third_party/py:python_init_rules.bzl", "python_init_rules")
+python_init_rules()
+
+load("@org_tensorflow//third_party/py:python_init_repositories.bzl", "python_init_repositories")
+python_init_repositories(
+    default_python_version = "system",
+    requirements = {
+        "3.10": "@org_tensorflow//:requirements_lock_3_10.txt",
+        "3.11": "@org_tensorflow//:requirements_lock_3_11.txt",
+        "3.12": "@org_tensorflow//:requirements_lock_3_12.txt",
+        "3.13": "@org_tensorflow//:requirements_lock_3_13.txt",
+    },
+)
+
+load("@org_tensorflow//third_party/py:python_init_toolchains.bzl", "python_init_toolchains")
+python_init_toolchains()
+
+load("@org_tensorflow//third_party/py:python_init_pip.bzl", "python_init_pip")
+python_init_pip()
+
+load("@pypi//:requirements.bzl", "install_deps")
+install_deps()
+
+# Initialize TensorFlow's external dependencies.
+load("@org_tensorflow//tensorflow:workspace3.bzl", "tf_workspace3")
+tf_workspace3()
+
+load("//tensorflow_serving:repo.bzl", "tf_serving_vendored")
+
+tf_serving_vendored(
+    name = "local_xla",
+    path = "third_party/xla",
+    repo_mapping = {
+        "@local_xla": "@local_xla",
+        "@local_tsl": "@local_tsl",
+        "@org_tensorflow": "@org_tensorflow",
+        "@xla": "@local_xla",
+        "@tsl": "@local_tsl",
+    },
+    root = "@org_tensorflow//:unused",
+)
+
+tf_serving_vendored(
+    name = "local_tsl",
+    path = "third_party/xla/third_party/tsl",
+    repo_mapping = {
+        "@local_xla": "@local_xla",
+        "@local_tsl": "@local_tsl",
+        "@org_tensorflow": "@org_tensorflow",
+        "@xla": "@local_xla",
+        "@tsl": "@local_tsl",
+    },
+    root = "@org_tensorflow//:unused",
+)
+load("@org_tensorflow//tensorflow:workspace2.bzl", "tf_workspace2")
+tf_workspace2()
+load("@org_tensorflow//tensorflow:workspace1.bzl", "tf_workspace1")
+tf_workspace1()
+load("@org_tensorflow//tensorflow:workspace0.bzl", "tf_workspace0")
+tf_workspace0()
+
+# Initialize bazel package rules' external dependencies.
+load("@rules_pkg//:deps.bzl", "rules_pkg_dependencies")
+rules_pkg_dependencies()
+
+load("@bazel_tools//tools/build_defs/repo:git.bzl", "git_repository")
+
+http_archive(
+    name = "rules_shell",
+    sha256 = "0d0c56d01c3c40420bf7bf14d73113f8a92fbd9f5cd13205a3b89f72078f0321",
+    strip_prefix = "rules_shell-0.1.1",
+    urls = [
+        "https://github.com/bazelbuild/rules_shell/releases/download/v0.1.1/rules_shell-v0.1.1.tar.gz",
+    ],
+)
+
+load("@rules_proto//proto:repositories.bzl", "rules_proto_dependencies")
+
+rules_proto_dependencies()
+
+load(
+    "@xla//third_party/py:python_wheel.bzl",
+    "python_wheel_version_suffix_repository",
+)
+
+python_wheel_version_suffix_repository(name = "tf_wheel_version_suffix")
+
+load(
+    "@rules_ml_toolchain//gpu/cuda:cuda_json_init_repository.bzl",
+    "cuda_json_init_repository",
+)
+
+cuda_json_init_repository()
+
+load(
+    "@cuda_redist_json//:distributions.bzl",
+    "CUDA_REDISTRIBUTIONS",
+    "CUDNN_REDISTRIBUTIONS",
+)
+load(
+    "@rules_ml_toolchain//gpu/cuda:cuda_redist_init_repositories.bzl",
+    "cuda_redist_init_repositories",
+    "cudnn_redist_init_repository",
+)
+
+cuda_redist_init_repositories(
+    cuda_redistributions = CUDA_REDISTRIBUTIONS,
+)
+
+cudnn_redist_init_repository(
+    cudnn_redistributions = CUDNN_REDISTRIBUTIONS,
+)
+
+load(
+    "@rules_ml_toolchain//gpu/cuda:cuda_configure.bzl",
+    "cuda_configure",
+)
+
+cuda_configure(name = "local_config_cuda")
+
+load(
+    "@rules_ml_toolchain//gpu/nccl:nccl_redist_init_repository.bzl",
+    "nccl_redist_init_repository",
+)
+
+nccl_redist_init_repository()
+
+load(
+    "@rules_ml_toolchain//gpu/nccl:nccl_configure.bzl",
+    "nccl_configure",
+)
+
+nccl_configure(name = "local_config_nccl")
+
+load(
+    "@rules_ml_toolchain//gpu/nvshmem:nvshmem_json_init_repository.bzl",
+    "nvshmem_json_init_repository",
+)
+
+nvshmem_json_init_repository()
+
+load(
+    "@nvshmem_redist_json//:distributions.bzl",
+    "NVSHMEM_REDISTRIBUTIONS",
+)
+load(
+    "@rules_ml_toolchain//gpu/nvshmem:nvshmem_redist_init_repository.bzl",
+    "nvshmem_redist_init_repository",
+)
+
+nvshmem_redist_init_repository(
+    nvshmem_redistributions = NVSHMEM_REDISTRIBUTIONS,
+)
+
+# nvshmem_configure removed in newer rules_ml_toolchain
+
diff --git a/tensorflow_serving/tools/docker/Dockerfile.devel b/tensorflow_serving/tools/docker/Dockerfile.devel
index 7c79a983f49..9f9286ecf7a 100644
--- a/tensorflow_serving/tools/docker/Dockerfile.devel
+++ b/tensorflow_serving/tools/docker/Dockerfile.devel
@@ -101,7 +101,7 @@ RUN echo "Building with build options: ${TF_SERVING_BUILD_OPTIONS}"
 ARG TF_SERVING_BAZEL_OPTIONS=""
 RUN echo "Building with Bazel options: ${TF_SERVING_BAZEL_OPTIONS}"
 
-RUN bazel build --color=yes --curses=yes \
+RUN bazel build --color=yes --noshow_progress --curses=no \
     ${TF_SERVING_BAZEL_OPTIONS} \
     --verbose_failures \
     --output_filter=DONT_MATCH_ANYTHING \
@@ -111,7 +111,7 @@ RUN bazel build --color=yes --curses=yes \
     /usr/local/bin/
 
 # Build and install TensorFlow Serving API
-RUN bazel build --color=yes --curses=yes \
+RUN bazel build --color=yes --noshow_progress --curses=no \
     ${TF_SERVING_BAZEL_OPTIONS} \
     --verbose_failures \
     --output_filter=DONT_MATCH_ANYTHING \
diff --git a/tensorflow_serving/tools/docker/Dockerfile.devel-gpu b/tensorflow_serving/tools/docker/Dockerfile.devel-gpu
index cde11aa6fe4..e914bc214ef 100644
--- a/tensorflow_serving/tools/docker/Dockerfile.devel-gpu
+++ b/tensorflow_serving/tools/docker/Dockerfile.devel-gpu
@@ -11,7 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-FROM nvidia/cuda:12.5.1-base-ubuntu22.04 as base_build
+FROM nvidia/cuda:12.2.2-base-ubuntu22.04 as base_build
 
 ARG TF_SERVING_VERSION_GIT_BRANCH=master
 ARG TF_SERVING_VERSION_GIT_COMMIT=HEAD
@@ -23,7 +23,11 @@ LABEL tensorflow_serving_github_commit=${TF_SERVING_VERSION_GIT_COMMIT}
 
 ENV CUDNN_VERSION=9.3.0.75
 ENV TF_TENSORRT_VERSION=8.6.1
-ENV CUDA=12.5
+ENV CUDA=12.2
+ENV HERMETIC_CUDA_VERSION=12.2.0
+
+ARG TF_CUDA_COMPUTE_CAPABILITIES="6.0,7.0,7.5,8.0,8.6,8.9,9.0"
+ENV TF_CUDA_COMPUTE_CAPABILITIES=${TF_CUDA_COMPUTE_CAPABILITIES}
 
 RUN apt-get update && apt-get install -y --no-install-recommends \
         curl
@@ -35,21 +39,21 @@ RUN curl -sSL --retry 5 https://github.com/tensorflow/serving/tarball/${TF_SERVI
 RUN /tensorflow-serving/tensorflow_serving/tools/docker/setup.sources.sh
 
 RUN apt-get install -y --no-install-recommends \
-        cuda-command-line-tools-12-5 \
-        cuda-cudart-dev-12-5 \
-        cuda-nvcc-12-5 \
-        cuda-cupti-12-5 \
-        cuda-nvprune-12-5 \
-        cuda-libraries-12-5 \
-        cuda-libraries-dev-12-5 \
-        cuda-nvml-dev-12-5 \
-        libcufft-12-5 \
-        libcurand-12-5 \
-        libcusolver-dev-12-5 \
-        libcusparse-dev-12-5 \
-        libcublas-12-5 \
-        libcublas-dev-12-5 \
-        cuda-compat-12-5 \
+        cuda-command-line-tools-12-2 \
+        cuda-cudart-dev-12-2 \
+        cuda-nvcc-12-2 \
+        cuda-cupti-12-2 \
+        cuda-nvprune-12-2 \
+        cuda-libraries-12-2 \
+        cuda-libraries-dev-12-2 \
+        cuda-nvml-dev-12-2 \
+        libcufft-12-2 \
+        libcurand-12-2 \
+        libcusolver-dev-12-2 \
+        libcusparse-dev-12-2 \
+        libcublas-12-2 \
+        libcublas-dev-12-2 \
+        cuda-compat-12-2 \
         libnccl-dev=2.25.1-1+cuda12.8 \
         libnccl2=2.25.1-1+cuda12.8 \
         libcudnn9-dev-cuda-12=9.3.0.75-1 \
@@ -150,12 +154,12 @@ FROM base_build as binary_build
 # Build, and install TensorFlow Serving
 ARG TF_SERVING_BUILD_OPTIONS="--config=release"
 RUN echo "Building with build options: ${TF_SERVING_BUILD_OPTIONS}"
-ARG TF_SERVING_BAZEL_OPTIONS=""
+ARG TF_SERVING_BAZEL_OPTIONS="--repo_env=HERMETIC_CUDA_VERSION=12.2.0 --repo_env=HERMETIC_CUDA_COMPUTE_CAPABILITIES=7.0,7.5,8.0,8.6,8.9,9.0"
 RUN echo "Building with Bazel options: ${TF_SERVING_BAZEL_OPTIONS}"
 
 RUN ln -s /usr/local/cuda/lib64/stubs/libcuda.so /usr/local/cuda/lib64/stubs/libcuda.so.1 && \
     LD_LIBRARY_PATH=/usr/local/cuda/lib64/stubs:${LD_LIBRARY_PATH} \
-    bazel build --color=yes --curses=yes --config=cuda_clang --copt="-fPIC"\
+    bazel build --color=yes --noshow_progress --curses=no --config=cuda_clang --copt="-fPIC"\
     ${TF_SERVING_BAZEL_OPTIONS} \
     --verbose_failures \
     --output_filter=DONT_MATCH_ANYTHING \
@@ -166,7 +170,7 @@ RUN ln -s /usr/local/cuda/lib64/stubs/libcuda.so /usr/local/cuda/lib64/stubs/lib
     rm /usr/local/cuda/lib64/stubs/libcuda.so.1
 
 # Build and install TensorFlow Serving API
-RUN bazel build --color=yes --curses=yes \
+RUN bazel build --color=yes --noshow_progress --curses=no \
     ${TF_SERVING_BAZEL_OPTIONS} \
     --verbose_failures \
     --output_filter=DONT_MATCH_ANYTHING \
diff --git a/tensorflow_serving/tools/docker/Dockerfile.devel-mkl b/tensorflow_serving/tools/docker/Dockerfile.devel-mkl
index 0a74fccc7a2..d87be197933 100644
--- a/tensorflow_serving/tools/docker/Dockerfile.devel-mkl
+++ b/tensorflow_serving/tools/docker/Dockerfile.devel-mkl
@@ -103,7 +103,7 @@ RUN echo "Building with build options: ${TF_SERVING_BUILD_OPTIONS}"
 ARG TF_SERVING_BAZEL_OPTIONS=""
 RUN echo "Building with Bazel options: ${TF_SERVING_BAZEL_OPTIONS}"
 
-RUN bazel build --color=yes --curses=yes \
+RUN bazel build --color=yes --noshow_progress --curses=no \
     ${TF_SERVING_BAZEL_OPTIONS} \
     --verbose_failures \
     --output_filter=DONT_MATCH_ANYTHING \
@@ -113,7 +113,7 @@ RUN bazel build --color=yes --curses=yes \
     /usr/local/bin/
 
 # Build and install TensorFlow Serving API
-RUN bazel build --color=yes --curses=yes \
+RUN bazel build --color=yes --noshow_progress --curses=no \
     ${TF_SERVING_BAZEL_OPTIONS} \
     --verbose_failures \
     --output_filter=DONT_MATCH_ANYTHING \
diff --git a/tensorflow_serving/tools/docker/Dockerfile.gpu b/tensorflow_serving/tools/docker/Dockerfile.gpu
index cee40ca9c9d..61389cce562 100644
--- a/tensorflow_serving/tools/docker/Dockerfile.gpu
+++ b/tensorflow_serving/tools/docker/Dockerfile.gpu
@@ -16,7 +16,7 @@ ARG TF_SERVING_VERSION=latest
 ARG TF_SERVING_BUILD_IMAGE=tensorflow/serving:${TF_SERVING_VERSION}-devel-gpu
 
 FROM ${TF_SERVING_BUILD_IMAGE} as build_image
-FROM nvidia/cuda:12.5.1-base-ubuntu22.04 as base_build
+FROM nvidia/cuda:12.2.2-base-ubuntu22.04 as base_build
 
 ARG TF_SERVING_VERSION_GIT_BRANCH=master
 ARG TF_SERVING_VERSION_GIT_COMMIT=HEAD
@@ -28,7 +28,8 @@ LABEL tensorflow_serving_github_commit=${TF_SERVING_VERSION_GIT_COMMIT}
 
 ENV CUDNN_VERSION=9.3.0.75
 ENV TF_TENSORRT_VERSION=8.6.1
-ENV CUDA=12.5
+ENV CUDA=12.2
+ENV HERMETIC_CUDA_VERSION=12.2.0
 ENV LD_LIBRARY_PATH /usr/local/cuda/extras/CUPTI/lib64:$LD_LIBRARY_PATH
 
 RUN apt-get update && apt-get install -y --no-install-recommends \
@@ -38,22 +39,22 @@ RUN curl -sSL --retry 5 https://raw.githubusercontent.com/tensorflow/serving/${T
 
 RUN apt-get install -y --no-install-recommends \
         ca-certificates \
-        cuda-command-line-tools-12-5 \
-        cuda-cupti-12-5 \
-        cuda-libraries-12-5 \
-        cuda-nvcc-12-5 \
-        cuda-nvprune-12-5 \
-        cuda-nvrtc-12-5 \
-        libcublas-12-5 \
+        cuda-command-line-tools-12-2 \
+        cuda-cupti-12-2 \
+        cuda-libraries-12-2 \
+        cuda-nvcc-12-2 \
+        cuda-nvprune-12-2 \
+        cuda-nvrtc-12-2 \
+        libcublas-12-2 \
         libcudnn9-cuda-12 \
-        libcufft-12-5 \
-        libcurand-12-5 \
-        libcusolver-12-5 \
-        libcusparse-12-5 \
+        libcufft-12-2 \
+        libcurand-12-2 \
+        libcusolver-12-2 \
+        libcusparse-12-2 \
         libnccl2 \
         libnvinfer8 \
         libnvinfer-plugin8 \
-        cuda-compat-12-5 \
+        cuda-compat-12-2 \
         libgomp1 \
         build-essential \
         curl \
diff --git a/third_party/libevent/BUILD b/third_party/libevent/BUILD
index 4a6dc97805f..fae802152c7 100644
--- a/third_party/libevent/BUILD
+++ b/third_party/libevent/BUILD
@@ -51,18 +51,28 @@ lib_files = [
 
 genrule(
     name = "libevent-srcs",
+    srcs = glob(["**"]) + ["@sysroot_linux_x86_64//:sysroot"],
     outs = include_files + lib_files,
     cmd = "\n".join([
         "export INSTALL_DIR=$$(pwd)/$(@D)/libevent",
+        "export ORIG_DIR=$$(pwd) # force rebuild 1",
+        "export CC_PATH=$(CC)",
+        "if [[ $$CC_PATH != /* ]]; then CC_PATH=$$ORIG_DIR/$$CC_PATH; fi",
+        "export REAL_CLANG=$$(ls $$ORIG_DIR/external/llvm*_linux_x86_64/bin/clang | head -n 1)",
+        "export GCC_PATH=$$REAL_CLANG",
+        "export REAL_AR=$$(ls $$ORIG_DIR/external/llvm*_linux_x86_64/bin/llvm-ar | head -n 1)",
+        "export SYSROOT_DIR=$$(ls -d $$ORIG_DIR/external/sysroot_linux_x86_64_glibc_* | head -n 1)",
         "export TMP_DIR=$$(mktemp -d -t libevent.XXXXXX)",
         "mkdir -p $$TMP_DIR",
-        "cp -R $$(pwd)/external/com_github_libevent_libevent/* $$TMP_DIR",
+        "cp -R $$ORIG_DIR/external/com_github_libevent_libevent/* $$TMP_DIR",
         "cd $$TMP_DIR",
         "./autogen.sh",
-        "./configure --prefix=$$INSTALL_DIR CFLAGS=-fPIC CXXFLAGS=-fPIC --enable-shared=no --disable-openssl",
-        "make install",
+        "CC=$$CC_PATH AR=$$REAL_AR ./configure --prefix=$$INSTALL_DIR CFLAGS=\"-fPIC --sysroot=$$SYSROOT_DIR\" CXXFLAGS=\"-fPIC --sysroot=$$SYSROOT_DIR\" LDFLAGS=\"--sysroot=$$SYSROOT_DIR\" --enable-shared=no --disable-openssl || (cat config.log && false)",
+        "make install CC=$$CC_PATH AR=$$REAL_AR",
         "rm -rf $$TMP_DIR",
     ]),
+    local = 1,
+    toolchains = ["@bazel_tools//tools/cpp:current_cc_toolchain"],
 )
 
 cc_library(
diff --git a/third_party/tensorflow/tensorflow.patch b/third_party/tensorflow/tensorflow.patch
index 75aa4f5f2ff..e69de29bb2d 100644
--- a/third_party/tensorflow/tensorflow.patch
+++ b/third_party/tensorflow/tensorflow.patch
@@ -1,10 +0,0 @@
-diff --git a/tensorflow/tools/toolchains/python/python_repo.bzl b/tensorflow/tools/toolchains/python/python_repo.bzl
---- a/tensorflow/tools/toolchains/python/python_repo.bzl
-+++ b/tensorflow/tools/toolchains/python/python_repo.bzl
-@@ -21,5 +21,6 @@
- TF_PYTHON_VERSION = "{}"
- HERMETIC_PYTHON_VERSION = "{}"
- WHEEL_NAME = "{}"
- WHEEL_COLLAB = "{}"
-+USE_PYWRAP_RULES = "False"
- """