diff --git a/.github/workflows/deploy.yml b/.github/workflows/deploy.yml
new file mode 100644
index 0000000000..aad4b036b4
--- /dev/null
+++ b/.github/workflows/deploy.yml
@@ -0,0 +1,26 @@
+name: deploy
+on: [push, pull_request, workflow_dispatch]
+jobs:
+  deploy:
+    name: Test Deployment (GCC ${{ matrix.gcc }})
+    runs-on: ubuntu-latest
+    container: nvidia/cuda:12.9.1-cudnn-devel-ubuntu24.04
+
+    steps:
+      - name: Checkout Git Repository
+        uses: actions/checkout@v3
+
+      - name: Install Depedencies
+        run: |
+          apt-get update -qq
+          apt-get install -y build-essential cmake curl gcc-${{ matrix.gcc }} g++-${{ matrix.gcc }} git libibverbs-dev libmpich-dev mpich python3 python3-venv zlib1g-dev
+
+      - name: Run Deploy Script
+        run: ./deploy/sapling.sh
+        env:
+          CC: gcc-${{ matrix.gcc }}
+          CXX: g++-${{ matrix.gcc }}
+
+    strategy:
+      matrix:
+        gcc: [10, 11]
diff --git a/.gitmodules b/.gitmodules
index 24bf52a686..e69de29bb2 100644
--- a/.gitmodules
+++ b/.gitmodules
@@ -1,24 +0,0 @@
-[submodule "deps/nccl"]
-	path = deps/nccl
-	url = https://github.com/NVIDIA/nccl.git
-[submodule "deps/json"]
-	path = deps/json
-	url = https://github.com/nlohmann/json.git
-[submodule "deps/spdlog"]
-	path = deps/spdlog
-	url = https://github.com/gabime/spdlog.git
-[submodule "deps/rapidcheck"]
-	path = deps/rapidcheck
-	url = https://github.com/emil-e/rapidcheck.git
-[submodule "deps/doctest"]
-	path = deps/doctest
-	url = https://github.com/doctest/doctest.git
-[submodule "deps/visit_struct"]
-	path = deps/visit_struct
-	url = https://github.com/cbeck88/visit_struct.git
-[submodule "deps/expected"]
-	path = deps/expected
-	url = https://github.com/TartanLlama/expected.git
-[submodule "deps/fmt"]
-	path = deps/fmt
-	url = https://github.com/fmtlib/fmt.git
diff --git a/.proj.toml b/.proj.toml
index 750f7e5955..522e11c369 100644
--- a/.proj.toml
+++ b/.proj.toml
@@ -111,6 +111,10 @@ cuda = false
 type = "bin"
 cuda = false
 
+[targets.run-model]
+type = "bin"
+cuda = false
+
 # default_build_targets = [
 #   "utils",
 # ]
@@ -124,5 +128,4 @@ cuda = false
 # ]
 
 [cmake_flags_extra]
-FF_CUDA_ARCH = "60"
 CMAKE_CUDA_ARCHITECTURES = "60"
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4723a3168d..4d62faf688 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,25 +1,13 @@
-cmake_minimum_required(VERSION 3.10)
+cmake_minimum_required(VERSION 3.18 FATAL_ERROR)
 project(FlexFlow)
 
-set(
-  CMAKE_MODULE_PATH 
-  ${CMAKE_MODULE_PATH} 
-  ${CMAKE_CURRENT_LIST_DIR}/cmake 
+list(
+  APPEND
+  CMAKE_MODULE_PATH
+  ${CMAKE_CURRENT_LIST_DIR}/cmake
   ${CMAKE_CURRENT_LIST_DIR}/cmake/Modules
 )
 
-# Detect OS type and Linux version (if it applies)
-set(LINUX_VERSION "")
-if(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
-  find_program(LSB_RELEASE_EXEC lsb_release)
-  if(LSB_RELEASE_EXEC)
-    execute_process(COMMAND ${LSB_RELEASE_EXEC} -r --short 
-                    OUTPUT_VARIABLE LINUX_VERSION 
-                    OUTPUT_STRIP_TRAILING_WHITESPACE)
-    message(STATUS "Linux Version: ${LINUX_VERSION}")
-  endif()
-endif()
-
 set(FF_MAX_DIM "5" CACHE STRING "Maximum tensor order")
 set(FF_MAX_OPNAME "128" CACHE STRING "Maximum op name length")
 set(FF_MAX_NUM_OUTPUTS "256" CACHE STRING "Maximum number of outputs (per operator)")
@@ -28,13 +16,12 @@ set(FF_MAX_NUM_WEIGHTS "64" CACHE STRING "Maximum number of weights (per operato
 set(FF_MAX_NUM_FUSED_OPERATORS "64" CACHE STRING "Maximum number of fused tensors")
 set(FF_MAX_NUM_FUSED_TENSORS "64" CACHE STRING "Maximum number of input and output tensors per fused op")
 set(FF_MAX_NUM_WORKERS "1024" CACHE STRING "Maximum number of GPUs")
-set(FF_MAX_NUM_TASK_REGIONS "20" CACHE STRING 
+set(FF_MAX_NUM_TASK_REGIONS "20" CACHE STRING
   "Maximum number of regions that can be passed to a task through the TaskSpec interface")
 set(FF_MAX_NUM_TASK_ARGUMENTS "5" CACHE STRING
   "Maximum number of arguments that can be declared in a TaskSignature")
 option(FF_USE_NCCL "Run FlexFlow with NCCL" OFF)
 option(FF_USE_PREALM "Build with PRealm profiling interface" ON)
-option(FF_USE_ALL_PREBUILT_LIBRARIES "Enable use of all pre-compiled libraries, if available" OFF)
 option(FF_USE_PYTHON "Enable Python" ON)
 option(FF_BUILD_FROM_PYPI "Build from pypi" OFF)
 option(FF_USE_CODE_COVERAGE "Enable code coverage" OFF)
@@ -43,16 +30,6 @@ set(FF_GPU_BACKENDS cuda hip_cuda hip_rocm intel)
 set(FF_GPU_BACKEND "cuda" CACHE STRING "Select GPU Backend ${FF_GPU_BACKENDS}")
 set_property(CACHE FF_GPU_BACKEND PROPERTY STRINGS ${FF_GPU_BACKENDS})
 
-option(FF_USE_EXTERNAL_NCCL "Use pre-installed NCCL" OFF)
-option(FF_USE_EXTERNAL_JSON "Use pre-installed nlohmann::json" OFF)
-option(FF_USE_EXTERNAL_FMT "Use pre-installed fmt" OFF)
-option(FF_USE_EXTERNAL_SPDLOG "Use pre-installed spdlog" OFF)
-option(FF_USE_EXTERNAL_DOCTEST "Use pre-installed doctest" OFF)
-option(FF_USE_EXTERNAL_RAPIDCHECK "Use pre-installed rapidcheck" OFF)
-option(FF_USE_EXTERNAL_EXPECTED "Use pre-installed tl::expected" OFF)
-option(FF_USE_EXTERNAL_GBENCHMARK "Use pre-installed google benchmark" OFF)
-option(FF_USE_EXTERNAL_LIBASSERT "Use pre-installed libassert" OFF)
-
 option(FF_BUILD_RESNET "build resnet example" OFF)
 option(FF_BUILD_RESNEXT "build resnext example" OFF)
 option(FF_BUILD_ALEXNET "build alexnet example" OFF)
@@ -72,15 +49,7 @@ option(FF_BUILD_VISUALIZATION_TOOL "build substitution visualization tool" ON)
 option(FF_BUILD_SP_IZATION_BENCHMARKING "build sp-ization benchmarking" ON)
 option(FF_BUILD_ARG_PARSER "build command line argument parser" OFF)
 option(FF_BUILD_BIN_EXPORT_MODEL_ARCH "build export-model-arch utility" ON)
-
-set(FF_CUDA_ARCH "autodetect" CACHE STRING "Target CUDA Arch")
-if (FF_CUDA_ARCH STREQUAL "")
-  message(FATAL_ERROR "FF_CUDA_ARCH cannot be an empty string. Set it to `autodetect`, `all`, or pass one or multiple valid CUDA archs.")
-endif()
-
-if(${CMAKE_SYSTEM_NAME} MATCHES "Linux")
-  set(LIBEXT ".so")
-endif()
+option(FF_BUILD_BIN_RUN_MODEL "build run-model binary" ON)
 
 include(cuda)
 include(cudnn)
diff --git a/bin/CMakeLists.txt b/bin/CMakeLists.txt
index 6855537460..7bf7909320 100644
--- a/bin/CMakeLists.txt
+++ b/bin/CMakeLists.txt
@@ -13,3 +13,7 @@ endif()
 if(FF_BUILD_BIN_EXPORT_MODEL_ARCH)
   add_subdirectory(export-model-arch)
 endif()
+
+if(FF_BUILD_BIN_RUN_MODEL)
+  add_subdirectory(run-model)
+endif()
diff --git a/bin/run-model/CMakeLists.txt b/bin/run-model/CMakeLists.txt
new file mode 100644
index 0000000000..a4e3f0ea9c
--- /dev/null
+++ b/bin/run-model/CMakeLists.txt
@@ -0,0 +1,10 @@
+ff_add_executable(
+  NAME
+    run-model
+  SRC_PATTERNS
+    src/*.cc
+  PRIVATE_INCLUDE
+    include/
+  DEPS
+    realm-execution
+)
diff --git a/bin/run-model/src/run-model/main.cc b/bin/run-model/src/run-model/main.cc
new file mode 100644
index 0000000000..46e1186b6b
--- /dev/null
+++ b/bin/run-model/src/run-model/main.cc
@@ -0,0 +1,117 @@
+#include "pcg/file_format/v1/v1_mapped_parallel_computation_graph.h"
+#include "pcg/mapped_parallel_computation_graph/mapped_parallel_computation_graph.dtg.h"
+#include "realm-execution/distributed_ff_handle.h"
+#include "realm-execution/pcg_instance.h"
+#include "realm-execution/realm_context.h"
+#include "realm-execution/realm_manager.h"
+#include "utils/cli/cli_get_help_message.h"
+#include "utils/cli/cli_parse.h"
+#include "utils/cli/cli_parse_result.h"
+#include "utils/cli/cli_spec.h"
+#include "utils/nonnegative_int/nonnegative_int.h"
+#include "utils/positive_int/positive_int.h"
+#include <fstream>
+#include <string_view>
+
+using namespace FlexFlow;
+
+static char *leak_string_contents(std::string_view str) {
+  // Realm command-line arguments require char* so intentionally leak the
+  // allocated string contents here
+  std::vector<char> *content = new std::vector<char>{str.begin(), str.end()};
+  content->push_back(0); // NUL byte
+  return content->data();
+}
+
+static std::vector<char *> make_realm_args(std::string_view executable_name) {
+  std::vector<char *> result;
+  result.push_back(leak_string_contents(executable_name));
+  return result;
+}
+
+int main(int argc, char **argv) {
+  CLISpec cli = empty_cli_spec();
+
+  CLIArgumentKey arg_key_help = cli_add_help_flag(cli);
+
+  CLIArgumentKey key_mapped_pcg_json = cli_add_positional_argument(
+      cli,
+      CLIPositionalArgumentSpec{
+          "mapped_pcg_json",
+          std::nullopt,
+          "path to a file containing mappped PCG encoded as JSON"});
+
+  ASSERT(argc >= 1);
+  std::string prog_name = argv[0];
+
+  CLIParseResult parsed = ({
+    tl::expected<CLIParseResult, std::string> result =
+        cli_parse(cli, argc, argv);
+    if (!result.has_value()) {
+      std::string error_msg = result.error();
+      std::cerr << cli_get_help_message(prog_name, cli);
+      std::cerr << std::endl;
+      std::cerr << "error: " << error_msg << std::endl;
+      return 1;
+    }
+
+    result.value();
+  });
+
+  bool help = cli_get_flag(parsed, arg_key_help);
+  if (help) {
+    std::cerr << cli_get_help_message(prog_name, cli);
+    return 1;
+  }
+
+  std::string mapped_pcg_json = cli_get_argument(parsed, key_mapped_pcg_json);
+
+  std::vector<char *> realm_args = make_realm_args(prog_name);
+  int realm_argc = realm_args.size();
+  char **realm_argv = realm_args.data();
+  RealmManager manager(&realm_argc, &realm_argv);
+
+  ControllerTaskResult result = manager.start_controller([&](RealmContext
+                                                                 &ctx) {
+    MappedParallelComputationGraph mpcg = [&]() {
+      std::ifstream f(mapped_pcg_json);
+      nlohmann::json mpcg_json = nlohmann::json::parse(f);
+      return from_v1(mpcg_json.get<V1MappedParallelComputationGraph>());
+    }();
+
+    // instantiate computation graph
+    OptimizerAttrs optimizer_attrs =
+        OptimizerAttrs{SGDOptimizerAttrs{/*lr=*/0.001,
+                                         /*momentum=*/0.9,
+                                         /*nesterov=*/false,
+                                         /*weight_decay=*/0.001}};
+
+    std::unordered_map<DynamicValueAttrs, DynamicTensorAccessor> input_tensors;
+
+    DistributedFfHandle device_handle =
+        create_distributed_ff_handle(ctx,
+                                     /*workSpaceSize=*/1024 * 1024,
+                                     /*allowTensorOpMathConversion=*/true);
+
+    PCGInstance pcg_instance = create_pcg_instance(
+        /*ctx=*/ctx,
+        /*mpcg=*/mpcg,
+        /*optimizer=*/optimizer_attrs,
+        /*loss=*/std::nullopt,
+        /*input_tensors=*/input_tensors,
+        /*profiling_settings=*/ProfilingSettings{0, 0},
+        /*device_handle=*/device_handle);
+
+    // begin training loop
+    int num_epochs = 5;
+    for (int i = 0; i < num_epochs; i++) {
+      perform_all_passes_for_pcg_instance(
+          /*instance=*/pcg_instance,
+          /*profiling_settings=*/ProfilingSettings{0, 1},
+          /*device_handle=*/device_handle);
+    }
+  });
+  result.wait();
+
+  return 0;
+}
diff --git a/bin/sp-ization-benchmarking/CMakeLists.txt b/bin/sp-ization-benchmarking/CMakeLists.txt
index c211ec067a..b2f2408d30 100644
--- a/bin/sp-ization-benchmarking/CMakeLists.txt
+++ b/bin/sp-ization-benchmarking/CMakeLists.txt
@@ -5,6 +5,8 @@ ff_add_executable(
     include/
   SRC_PATTERNS
     *.cc
+  PRIVATE_INCLUDE
+    include/
   DEPS
     utils
     rapidcheck
diff --git a/cmake/Modules/CodeCoverage.cmake b/cmake/Modules/CodeCoverage.cmake
index 2556b19f7b..5a8fe67057 100644
--- a/cmake/Modules/CodeCoverage.cmake
+++ b/cmake/Modules/CodeCoverage.cmake
@@ -161,7 +161,7 @@ foreach(LANG ${LANGUAGES})
     endif()
   elseif(NOT "${CMAKE_${LANG}_COMPILER_ID}" MATCHES "GNU"
          AND NOT "${CMAKE_${LANG}_COMPILER_ID}" MATCHES "(LLVM)?[Ff]lang")
-    message(FATAL_ERROR "Compiler is not GNU or Flang! Aborting...")
+    message(FATAL_ERROR "Compiler ${CMAKE_${LANG}_COMPILER_ID} is not GNU or Flang! Aborting...")
   endif()
 endforeach()
 
@@ -748,4 +748,4 @@ function(append_coverage_compiler_flags_to_target name)
     if(CMAKE_C_COMPILER_ID STREQUAL "GNU" OR CMAKE_CXX_COMPILER_ID STREQUAL "GNU" OR CMAKE_Fortran_COMPILER_ID STREQUAL "GNU")
         target_link_libraries(${name} PRIVATE gcov)
     endif()
-endfunction()
\ No newline at end of file
+endfunction()
diff --git a/cmake/cuda.cmake b/cmake/cuda.cmake
index da358e31dd..e12ac52589 100644
--- a/cmake/cuda.cmake
+++ b/cmake/cuda.cmake
@@ -1,89 +1,5 @@
-set(CUDA_USE_STATIC_CUDA_RUNTIME OFF)
+include(aliasing)
 
-set(CUDA_ROOT ${CUDA_PATH})
-set(CUDA_TOOLKIT_ROOT_DIR ${CUDA_PATH})
-list(APPEND CMAKE_PREFIX_PATH ${CUDA_ROOT})
-find_package(CUDA REQUIRED)
-message("cuda-libraries ${CUDA_LIBRARIES}")
+find_package(CUDAToolkit 11.7 REQUIRED)
 
-if(CUDA_FOUND)
-  # strip the cudart lib
-  string(REGEX REPLACE "[^\;]*cudart[^\;]*(\;?)" "" CUDA_LIBRARIES "${CUDA_LIBRARIES}")
-  set(CUDA_LIBRARIES ${CUDA_LIBRARIES})
-
-  # set cuda runtime and driver lib
-  # override cublas and curand because the FindCUDA module may not find the correct libs  
-  set(CUDADRV_LIBRARIES ${CUDA_TOOLKIT_ROOT_DIR}/lib64/stubs/libcuda${LIBEXT})
-  set(CUDA_CUBLAS_LIBRARIES ${CUDA_TOOLKIT_ROOT_DIR}/lib64/libcublas${LIBEXT})
-  set(CUDA_curand_LIBRARY ${CUDA_TOOLKIT_ROOT_DIR}/lib64/libcurand${LIBEXT})
-  list(APPEND FLEXFLOW_EXT_LIBRARIES
-    ${CUDADRV_LIBRARIES}
-    ${CUDA_CUBLAS_LIBRARIES}
-    ${CUDA_curand_LIBRARY})
-
-  # Snippet below from legion/cmake/newcmake/FindCUDA.cmake
-  # Find the `nvcc` executable
-  find_program(CUDA_NVCC_EXECUTABLE
-    NAMES nvcc
-    PATHS "${CUDA_TOOLKIT_ROOT_DIR}"
-    ENV CUDA_PATH
-    ENV CUDA_BIN_PATH
-    PATH_SUFFIXES bin bin64
-    NO_DEFAULT_PATH
-  )
-  # Search default search paths, after we search our own set of paths.
-  find_program(CUDA_NVCC_EXECUTABLE nvcc)
-  mark_as_advanced(CUDA_NVCC_EXECUTABLE)
-  # Compute the CUDA version.
-  if(CUDA_NVCC_EXECUTABLE AND NOT CUDA_VERSION)
-    execute_process (COMMAND ${CUDA_NVCC_EXECUTABLE} "--version" OUTPUT_VARIABLE NVCC_OUT)
-    string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\1" CUDA_VERSION_MAJOR ${NVCC_OUT})
-    string(REGEX REPLACE ".*release ([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR ${NVCC_OUT})
-    set(CUDA_VERSION "${CUDA_VERSION_MAJOR}.${CUDA_VERSION_MINOR}" CACHE STRING "Version of CUDA as computed from nvcc.")
-    mark_as_advanced(CUDA_VERSION)
-  else()
-    # Need to set these based off of the cached value
-    string(REGEX REPLACE "([0-9]+)\\.([0-9]+).*" "\\1" CUDA_VERSION_MAJOR "${CUDA_VERSION}")
-    string(REGEX REPLACE "([0-9]+)\\.([0-9]+).*" "\\2" CUDA_VERSION_MINOR "${CUDA_VERSION}")
-  endif()
-
-  # Set FF_CUDA_ARCH to the list of GPU architectures found on the machine.
-  if("${FF_CUDA_ARCH}" STREQUAL "autodetect")
-    include(utils)
-    detect_installed_gpus(DETECTED_CUDA_ARCH)
-    message( STATUS "CUDA Detected CUDA_ARCH : ${DETECTED_CUDA_ARCH}" )
-    set(FF_CUDA_ARCH ${DETECTED_CUDA_ARCH})
-  # Set FF_CUDA_ARCH to the list of all GPU architectures compatible with FlexFlow
-  elseif("${FF_CUDA_ARCH}" STREQUAL "all") 
-    set(FF_CUDA_ARCH 60,61,62,70,72,75,80,86)
-  endif()
-  
-  # create CUDA_GENCODE list based on FF_CUDA_ARCH
-  string(REPLACE "," ";" CUDA_GENCODE "${FF_CUDA_ARCH}")
-  foreach(CODE ${CUDA_GENCODE})
-    if(CODE LESS 60)
-      message( FATAL_ERROR "CUDA architecture <60 not supported")
-    endif()
-  endforeach()
-  string(REGEX REPLACE "([0-9]+)" "-gencode arch=compute_\\1,code=sm_\\1" CUDA_GENCODE "${CUDA_GENCODE}")
-
-  #output
-  message( STATUS "CUDA_VERSION: ${CUDA_VERSION}")
-  message( STATUS "CUDA root path : ${CUDA_TOOLKIT_ROOT_DIR}" )
-  message( STATUS "CUDA include path : ${CUDA_INCLUDE_DIRS}" )
-  message( STATUS "CUDA runtime libraries : ${CUDA_LIBRARIES}" )
-  message( STATUS "CUDA driver libraries : ${CUDADRV_LIBRARIES}" )
-  message( STATUS "CUBLAS libraries : ${CUDA_CUBLAS_LIBRARIES}" )
-  message( STATUS "CURAND libraries : ${CUDA_curand_LIBRARY}" )
-  message( STATUS "CUDA Arch : ${FF_CUDA_ARCH}" )
-  message( STATUS "CUDA_GENCODE: ${CUDA_GENCODE}")
-
-  list(APPEND FLEXFLOW_INCLUDE_DIRS
-    ${CUDA_INCLUDE_DIRS})
-
-  add_library(cuda INTERFACE)
-  target_include_directories(cuda SYSTEM INTERFACE "${CUDA_INCLUDE_DIRS}")
-
-else()
-  message( FATAL_ERROR "CUDA package not found -> specify search path via CUDA_ROOT variable")
-endif()
+alias_library(deps::cublas CUDA::cublas)
diff --git a/cmake/cudnn.cmake b/cmake/cudnn.cmake
index 1055931cef..326fce5cc1 100644
--- a/cmake/cudnn.cmake
+++ b/cmake/cudnn.cmake
@@ -1,46 +1,16 @@
-# find cudnn in CUDNN_ROOT and CUDA_ROOT
-if(CUDNN_PATH)
-  set(CUDNN_ROOT ${CUDNN_PATH})
-else()
-	# if CUDNN_PATH is not set, let's try to find it in the CUDA root
-	set(CUDNN_ROOT ${CUDA_TOOLKIT_ROOT_DIR})
-	message( "CMAKE_LIBRARY_PATH ${CMAKE_LIBRARY_PATH}" )
-endif()
-find_library(CUDNN_LIBRARY 
-  NAMES cudnn
-  PATHS ${CUDNN_ROOT} ${CUDA_ROOT}
-  PATH_SUFFIXES lib lib64
-  DOC "CUDNN library." )
-  
-find_path(CUDNN_INCLUDE_DIR 
-    NAMES cudnn.h
-    HINTS ${CUDNN_ROOT} ${CUDA_ROOT}
-    PATH_SUFFIXES include 
-    DOC "CUDNN include directory." )
-
-# find cudnn, set cudnn lib and include    
-if(CUDNN_LIBRARY AND CUDNN_INCLUDE_DIR)
-  set(CUDNN_FOUND ON)
-  set(CUDNN_LIBRARIES ${CUDNN_LIBRARY})
-  set(CUDNN_INCLUDE_DIRS ${CUDNN_INCLUDE_DIR})
-endif()
+include(aliasing)
 
-# find cuda and cudnn
-if(CUDNN_FOUND)
-  list(APPEND FLEXFLOW_EXT_LIBRARIES
-    ${CUDNN_LIBRARIES})
+find_path(CUDNN_INCLUDE_DIR NAMES cudnn.h HINTS ${CUDAToolkit_INCLUDE_DIRS})
+find_library(CUDNN_LIBRARY NAMES cudnn HINTS ${CUDAToolkit_LIBRARY_DIR})
 
-  list(APPEND FLEXFLOW_INCLUDE_DIRS
-    ${CUDNN_INCLUDE_DIR})
-endif()
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(CUDNN DEFAULT_MSG CUDNN_LIBRARY CUDNN_INCLUDE_DIR)
 
 if(CUDNN_FOUND)
-message( STATUS "CUDNN include : ${CUDNN_INCLUDE_DIR}" )
-  message( STATUS "CUDNN libraries : ${CUDNN_LIBRARIES}" )
-else()
-  message( FATAL_ERROR "CUDNN package not found -> specify search path via CUDNN_DIR variable")
+  add_library(cudnn UNKNOWN IMPORTED)
+  set_target_properties(cudnn PROPERTIES
+    IMPORTED_LOCATION "${CUDNN_LIBRARY}"
+    INTERFACE_INCLUDE_DIRECTORIES "${CUDNN_INCLUDE_DIR}"
+  )
+  alias_library(deps::cudnn cudnn)
 endif()
-
-add_library(cudnn INTERFACE)
-target_include_directories(cudnn SYSTEM INTERFACE ${CUDNN_INCLUDE_DIR})
-target_link_libraries(cudnn INTERFACE ${CUDNN_LIBRARY})
diff --git a/cmake/doctestlib.cmake b/cmake/doctestlib.cmake
index 6a41a0d5f0..d6b5953cd4 100644
--- a/cmake/doctestlib.cmake
+++ b/cmake/doctestlib.cmake
@@ -1,16 +1,11 @@
 include(aliasing)
-  
-if (FF_USE_EXTERNAL_DOCTEST)
-  find_package(doctest REQUIRED)
-  include(doctest) # import doctest_discover_tests
 
-  target_compile_definitions(
-    doctest::doctest
-    INTERFACE
-      DOCTEST_CONFIG_REQUIRE_STRINGIFICATION_FOR_ALL_USED_TYPES
-  )
-  alias_library(doctest doctest::doctest)
-else()
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/deps/doctest)
-  include(${CMAKE_CURRENT_SOURCE_DIR}/deps/doctest/scripts/cmake/doctest.cmake)
-endif()
+find_package(doctest REQUIRED)
+include(doctest) # import doctest_discover_tests
+
+target_compile_definitions(
+  doctest::doctest
+  INTERFACE
+    DOCTEST_CONFIG_REQUIRE_STRINGIFICATION_FOR_ALL_USED_TYPES
+)
+alias_library(deps::doctest doctest::doctest)
diff --git a/cmake/expected.cmake b/cmake/expected.cmake
index 17c73532fe..86f6a69fcd 100644
--- a/cmake/expected.cmake
+++ b/cmake/expected.cmake
@@ -1,9 +1,3 @@
 include(aliasing)
-if (FF_USE_EXTERNAL_EXPECTED)
-  find_package(tl-expected REQUIRED)
-  alias_library(expected tl::expected)
-else()
-  set(EXPECTED_BUILD_TESTS OFF)
-  set(EXPECTED_BUILD_PACKAGE OFF)
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/deps/expected)
-endif()
+find_package(tl-expected REQUIRED)
+alias_library(deps::expected tl::expected)
diff --git a/cmake/flexflow-utils.cmake b/cmake/flexflow-utils.cmake
index 7d5d189b1c..5dccd5272d 100644
--- a/cmake/flexflow-utils.cmake
+++ b/cmake/flexflow-utils.cmake
@@ -113,8 +113,8 @@ function(ff_add_test_executable)
       DEPS
     PARSE
       ${ARGN}
-      rapidcheck
-      doctest
+      deps::rapidcheck
+      deps::doctest
   )
 
   project(${FF_TEST_EXEC_NAME})
@@ -180,8 +180,8 @@ function(ff_add_benchmark_executable)
   target_link_libraries(
     ${FF_BENCHMARK_EXEC_NAME}
     ${FF_BENCHMARK_EXEC_DEPS}
-    gbenchmark
-    gbenchmark-main)
+    deps::gbenchmark
+    deps::gbenchmark-main)
 
   define_ff_vars(${FF_BENCHMARK_EXEC_NAME})
   ff_set_cxx_properties(${FF_BENCHMARK_EXEC_NAME})
diff --git a/cmake/fmt.cmake b/cmake/fmt.cmake
index 470de6a847..59df33a52f 100644
--- a/cmake/fmt.cmake
+++ b/cmake/fmt.cmake
@@ -1,8 +1,3 @@
 include(aliasing)
-
-if (FF_USE_EXTERNAL_FMT)
-  find_package(fmt REQUIRED)
-else()
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/deps/fmt)
-endif()
-alias_library(fmt fmt::fmt)
+find_package(fmt REQUIRED)
+alias_library(deps::fmt fmt::fmt)
diff --git a/cmake/gbenchmark.cmake b/cmake/gbenchmark.cmake
index d5bab19a1f..30b820855f 100644
--- a/cmake/gbenchmark.cmake
+++ b/cmake/gbenchmark.cmake
@@ -1,10 +1,4 @@
 include(aliasing)
-
-if (FF_USE_EXTERNAL_GBENCHMARK)
-  find_package(benchmark REQUIRED)
-  alias_library(gbenchmark benchmark::benchmark)
-  alias_library(gbenchmark-main benchmark::benchmark_main)
-else()
-  message(FATAL_ERROR "Currently FF_USE_EXTERNAL_GBENCHMARK is required")
-endif()
-
+find_package(benchmark REQUIRED)
+alias_library(deps::gbenchmark benchmark::benchmark)
+alias_library(deps::gbenchmark-main benchmark::benchmark_main)
diff --git a/cmake/json.cmake b/cmake/json.cmake
index 093ec51cdc..658afcb82e 100644
--- a/cmake/json.cmake
+++ b/cmake/json.cmake
@@ -1,12 +1,5 @@
 include(aliasing)
 
-if (FF_USE_EXTERNAL_JSON)
-  find_package(nlohmann_json REQUIRED)
+find_package(nlohmann_json REQUIRED)
 
-  alias_library(json nlohmann_json)
-else()
-  set(JSON_BuildTests OFF CACHE INTERNAL "")
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/deps/json)
-
-  alias_library(json nlohmann_json::nlohmann_json)
-endif()
+alias_library(deps::json nlohmann_json)
diff --git a/cmake/libassert.cmake b/cmake/libassert.cmake
index 3abe03b014..f9290ee919 100644
--- a/cmake/libassert.cmake
+++ b/cmake/libassert.cmake
@@ -1,9 +1,5 @@
 include(aliasing)
 
-if(FF_USE_EXTERNAL_LIBASSERT)
-  find_package(libassert REQUIRED)
-else()
-  message(FATAL_ERROR "Currently FF_USE_EXTERNAL_LIBASSERT is required")
-endif()
+find_package(libassert REQUIRED)
 
-alias_library(libassert libassert::assert)
+alias_library(deps::libassert libassert::assert)
diff --git a/cmake/nccl.cmake b/cmake/nccl.cmake
index 755fe00f1b..d831dde1db 100644
--- a/cmake/nccl.cmake
+++ b/cmake/nccl.cmake
@@ -1,37 +1,8 @@
 include(aliasing)
 
-add_library(nccl INTERFACE)
-
-if (FF_USE_EXTERNAL_NCCL)
-  find_package(NCCL REQUIRED)
-else()
-  message(STATUS "Building NCCL from source")
-  list(TRANSFORM CUDA_GENCODE PREPEND "NVCC_GENCODE=" OUTPUT_VARIABLE NCCL_BUILD_NVCC_GENCODE)
-
-  include(ExternalProject)
-  ExternalProject_Add(nccl_source_build
-    SOURCE_DIR ${PROJECT_SOURCE_DIR}/deps/${NCCL_NAME}
-    PREFIX ${CMAKE_BINARY_DIR}/deps/${NCCL_NAME}
-    INSTALL_DIR ${CMAKE_BINARY_DIR}/deps/${NCCL_NAME}
-    BUILD_BYPRODUCTS ${CMAKE_BINARY_DIR}/deps/${NCCL_NAME}/lib/libnccl${LIBEXT}
-    INSTALL_COMMAND ""
-    CONFIGURE_COMMAND ""
-    BUILD_COMMAND make src.build "${NCCL_BUILD_NVCC_GENCODE}" "CUDA_HOME=${CUDA_TOOLKIT_ROOT_DIR}" "BUILDDIR=${CMAKE_BINARY_DIR}/deps/${NCCL_NAME}"
-    BUILD_IN_SOURCE 1
-  )
-
-  ExternalProject_Get_Property(nccl_source_build INSTALL_DIR)
-  set_directory_properties(PROPERTIES ADDITIONAL_CLEAN_FILES "${CMAKE_BINARY_DIR}/deps/nccl_source_build/lib/")
-  
-  install(DIRECTORY ${CMAKE_BINARY_DIR}/deps/nccl_source_build/include/ DESTINATION include)
-  install(DIRECTORY ${CMAKE_BINARY_DIR}/deps/nccl_source_build/lib/ DESTINATION lib PATTERN "pkgconfig" EXCLUDE)
+find_package(NCCL REQUIRED)
 
-  set(NCCL_INCLUDE_DIR "${INSTALL_DIR}/include")
-  set(NCCL_LIBRARIES "${INSTALL_DIR}/lib/libnccl${LIBEXT}")
-
-  add_dependencies(nccl nccl_source_build)
-endif()
-
-message(STATUS "NCCL_LIBRARIES = ${NCCL_LIBRARIES}")
+add_library(nccl INTERFACE)
 target_include_directories(nccl SYSTEM INTERFACE ${NCCL_INCLUDE_DIRS})
 target_link_libraries(nccl INTERFACE ${NCCL_LIBRARIES})
+alias_library(deps::nccl nccl)
diff --git a/cmake/rapidcheck.cmake b/cmake/rapidcheck.cmake
index bf8f058e63..8180cc7436 100644
--- a/cmake/rapidcheck.cmake
+++ b/cmake/rapidcheck.cmake
@@ -1,5 +1,5 @@
-if (FF_USE_EXTERNAL_RAPIDCHECK)
-  find_package(rapidcheck REQUIRED)
-else()
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/deps/rapidcheck)
-endif()
+include(aliasing)
+
+find_package(rapidcheck REQUIRED)
+
+alias_library(deps::rapidcheck rapidcheck)
diff --git a/cmake/realm.cmake b/cmake/realm.cmake
index 91d8f1345c..ace8f13f1c 100644
--- a/cmake/realm.cmake
+++ b/cmake/realm.cmake
@@ -2,4 +2,4 @@ include(aliasing)
 
 find_package(Realm REQUIRED)
 
-alias_library(realm Realm::Realm)
+alias_library(deps::realm Realm::Realm)
diff --git a/cmake/spdlog.cmake b/cmake/spdlog.cmake
index 5ba1d6cc15..3dbe0c8e95 100644
--- a/cmake/spdlog.cmake
+++ b/cmake/spdlog.cmake
@@ -1,12 +1,5 @@
 include(aliasing)
 
-if (FF_USE_EXTERNAL_SPDLOG)
-  find_package(spdlog REQUIRED)
-else()
-  add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/deps/spdlog)
-endif()
+find_package(spdlog REQUIRED)
 
-add_library(ff_spdlog INTERFACE)
-target_link_libraries(ff_spdlog INTERFACE spdlog::spdlog)
-target_compile_definitions(ff_spdlog INTERFACE SPDLOG_FMT_EXTERNAL)
-alias_library(spdlog ff_spdlog)
+alias_library(deps::spdlog spdlog::spdlog)
diff --git a/cmake/utils.cmake b/cmake/utils.cmake
deleted file mode 100644
index 4e23ed2e3f..0000000000
--- a/cmake/utils.cmake
+++ /dev/null
@@ -1,56 +0,0 @@
-set(known_gpu_archs "")
-function(remove_duplicate_args __string)
-  if(${__string})
-    set(__list ${${__string}})
-    separate_arguments(__list)
-    list(REMOVE_DUPLICATES __list)
-    foreach(__e ${__list})
-      set(__str "${__str} ${__e}")
-    endforeach()
-    set(${__string} ${__str} PARENT_SCOPE)
-  endif()
-endfunction()
-function(detect_installed_gpus out_variable)
-  if(NOT CUDA_gpu_detect_output)
-    set(__cufile ${PROJECT_BINARY_DIR}/detect_cuda_archs.cu)
-    file(WRITE ${__cufile} ""
-      "#include <cstdio>\n"
-      "int main()\n"
-      "{\n"
-      "  int count = 0;\n"
-      "  if (cudaSuccess != cudaGetDeviceCount(&count)) return -1;\n"
-      "  if (count == 0) return -1;\n"
-      "  for (int device = 0; device < count; ++device)\n"
-      "  {\n"
-      "    cudaDeviceProp prop;\n"
-      "    if (cudaSuccess == cudaGetDeviceProperties(&prop, device))\n"
-      "      std::printf(\"%d.%d \", prop.major, prop.minor);\n"
-      "  }\n"
-      "  return 0;\n"
-      "}\n")
-    execute_process(COMMAND "${CUDA_NVCC_EXECUTABLE}" "--run" "${__cufile}"
-                    WORKING_DIRECTORY "${PROJECT_BINARY_DIR}/CMakeFiles/"
-                    RESULT_VARIABLE __nvcc_res OUTPUT_VARIABLE __nvcc_out
-                    ERROR_QUIET OUTPUT_STRIP_TRAILING_WHITESPACE)
-    if(__nvcc_res EQUAL 0)
-      message(STATUS "No result from nvcc so building for 2.0")
-      string(REPLACE "2.1" "2.1(2.0)" __nvcc_out "${__nvcc_out}")
-      set(CUDA_gpu_detect_output ${__nvcc_out} CACHE INTERNAL "Returned GPU architetures from detect_gpus tool" FORCE)
-    endif()
-  endif()
-  if(NOT CUDA_gpu_detect_output)
-    message(STATUS "Automatic GPU detection failed, Architecture is not set: ${known_gpu_archs}.")
-    set(${out_variable} ${known_gpu_archs} PARENT_SCOPE)
-  else()
-    remove_duplicate_args(CUDA_gpu_detect_output)
-    #Strip leading and trailing whitespaces
-    string(STRIP "${CUDA_gpu_detect_output}" CUDA_gpu_detect_output)
-    #Replace spaces in between with commas so you go from "5.2 6.1" to "5.2,6.1"
-    string(REGEX REPLACE " " "," CUDA_gpu_detect_output "${CUDA_gpu_detect_output}")
-    # message(${CUDA_gpu_detect_output})
-    string(REPLACE "." "" CUDA_gpu_detect_output "${CUDA_gpu_detect_output}")
-    # message(${CUDA_gpu_detect_output})
-    set(${out_variable} ${CUDA_gpu_detect_output} PARENT_SCOPE)
-    # message(STATUS "Automatic GPU ARCH detection: ${CUDA_gpu_detect_output}")
-  endif()
-endfunction()
diff --git a/cmake/zlib.cmake b/cmake/zlib.cmake
deleted file mode 100644
index 0281e02b88..0000000000
--- a/cmake/zlib.cmake
+++ /dev/null
@@ -1,8 +0,0 @@
-find_package(ZLIB REQUIRED)
-if(ZLIB_FOUND)
-  list(APPEND FLEXFLOW_EXT_LIBRARIES 
-    ${ZLIB_LIBRARIES})
-  message( STATUS "ZLIB libraries : ${ZLIB_LIBRARIES}" )
-else()
-  message( FATAL_ERROR "ZLIB package not found")
-endif()
\ No newline at end of file
diff --git a/deploy/sapling.sh b/deploy/sapling.sh
new file mode 100755
index 0000000000..04d43e252d
--- /dev/null
+++ b/deploy/sapling.sh
@@ -0,0 +1,147 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+# These modules are specific to Sapling, if deploying to another machine
+# customize as necessary.
+if [[ ! -v CI ]]; then
+    module load cuda cmake
+fi
+
+err_out() {
+    >&2 echo "$@"
+    exit 1
+}
+
+while [[ $# -gt 0 ]]; do
+  case $1 in
+    --jobs|-j)
+      THREADS="$2"
+      shift
+      shift
+      ;;
+    -*)
+      err_out "Unknown option: $1"
+      ;;
+    *)
+      err_out "Unknown argument: $1"
+      ;;
+  esac
+done
+
+: "${CC:=gcc-10}"
+: "${CXX:=g++-10}"
+: "${CUDAARCHS:=60}"
+: "${THREADS:=$(nproc)}"
+: "${CMAKE_PREFIX_PATH:=}"
+
+export CC
+export CXX
+export CUDAARCHS
+export THREADS
+
+mkdir -p deploy/deps
+pushd deploy/deps
+
+require_cmd() {
+    local cmd="$1"
+    if ! command -v "$cmd" >/dev/null 2>&1; then
+        err_out "Could not find command: $cmd"
+    fi
+}
+
+require_cmd "$CC"
+require_cmd "$CXX"
+require_cmd cmake
+require_cmd curl
+require_cmd git
+require_cmd make
+require_cmd nvcc
+require_cmd python3
+require_cmd tar
+
+function build_cmake_library {
+    dep_name="$1"
+    dep_url="$2"
+    dep_args=("${@:3}")
+    if [[ ! -e ${dep_name} ]]; then
+        if [[ ${dep_url} == *.git ]]; then
+            git clone --depth 1 --single-branch "${dep_url}" "${dep_name}"
+        else
+            mkdir "${dep_name}"
+            curl -LsSf -o "${dep_name}.tar.gz" "${dep_url}"
+            tar xfz "${dep_name}.tar.gz" -C "${dep_name}" --strip-components=1
+        fi
+    fi
+    if [[ ! -e "${dep_name}"_install/include ]]; then
+        mkdir -p "${dep_name}"_build "${dep_name}"_install
+        pushd "${dep_name}"_build
+        cmake ../"${dep_name}" -DCMAKE_INSTALL_PREFIX="$PWD"/../"${dep_name}"_install "${dep_args[@]}"
+        make install -j"$THREADS"
+        popd
+        rm -rf "${dep_name}"_build
+    fi
+    export CMAKE_PREFIX_PATH="$CMAKE_PREFIX_PATH:$PWD/${dep_name}_install"
+}
+
+if [[ ! -e gasnet ]]; then
+    git clone --depth 1 --single-branch https://github.com/StanfordLegion/gasnet.git
+fi
+if [[ ! -e gasnet/release ]]; then
+    make -C gasnet CONDUIT=ibv
+fi
+export GASNet_ROOT="$PWD"/gasnet/release
+
+set -x
+
+build_cmake_library zstd https://github.com/facebook/zstd.git -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+
+build_cmake_library fmt https://github.com/fmtlib/fmt/archive/refs/tags/10.2.1.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+
+build_cmake_library cpptrace https://github.com/jeremy-rifkin/cpptrace/archive/refs/tags/v1.0.4.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DCPPTRACE_USE_EXTERNAL_ZSTD=ON
+
+build_cmake_library libassert https://github.com/jeremy-rifkin/libassert/archive/refs/tags/v2.2.1.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DLIBASSERT_USE_EXTERNAL_CPPTRACE=ON
+
+build_cmake_library Realm https://github.com/StanfordLegion/realm.git -DCMAKE_BUILD_TYPE=RelWithDebInfo -DBUILD_SHARED_LIBS=ON -DREALM_ENABLE_CUDA=ON -DREALM_ENABLE_PREALM=ON -DREALM_ENABLE_CPPTRACE=ON -DREALM_ENABLE_HDF5=OFF -DREALM_MAX_DIM=5
+
+build_cmake_library benchmark https://github.com/google/benchmark/archive/refs/tags/v1.9.5.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DBENCHMARK_DOWNLOAD_DEPENDENCIES=ON
+
+build_cmake_library rapidcheck https://github.com/emil-e/rapidcheck.git -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+
+build_cmake_library tl-expected https://github.com/TartanLlama/expected/archive/refs/tags/v1.3.1.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+
+build_cmake_library doctest https://github.com/doctest/doctest/archive/refs/tags/v2.4.12.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+
+build_cmake_library spdlog https://github.com/gabime/spdlog/archive/refs/tags/v1.17.0.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON -DSPDLOG_FMT_EXTERNAL=ON
+
+build_cmake_library nlohmann_json https://github.com/nlohmann/json/archive/refs/tags/v3.12.0.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+
+build_cmake_library NCCL https://github.com/NVIDIA/nccl/archive/refs/tags/v2.29.7-1.tar.gz -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+
+mkdir -p proj/
+if [[ ! -e proj/venv ]]; then
+    python3 -m venv proj/venv
+fi
+
+# shellcheck disable=SC1091
+source proj/venv/bin/activate
+
+if ! command -v proj >/dev/null 2>&1
+then
+    pip install --require-virtualenv 'git+https://git.sr.ht/~lockshaw/proj'
+fi
+
+popd # deploy/deps
+
+ff_cmake_flags=(
+    -DCMAKE_BUILD_TYPE=RelWithDebInfo
+    -DCMAKE_INSTALL_PREFIX="$PWD/../install"
+)
+
+proj dtgen
+
+mkdir build install
+pushd build
+cmake .. "${ff_cmake_flags[@]}"
+make -j"$THREADS"
+popd # build
diff --git a/deps/doctest b/deps/doctest
deleted file mode 160000
index b7c21ec5ce..0000000000
--- a/deps/doctest
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit b7c21ec5ceeadb4951b00396fc1e4642dd347e5f
diff --git a/deps/expected b/deps/expected
deleted file mode 160000
index 292eff8bd8..0000000000
--- a/deps/expected
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit 292eff8bd8ee230a7df1d6a1c00c4ea0eb2f0362
diff --git a/deps/fmt b/deps/fmt
deleted file mode 160000
index f5e54359df..0000000000
--- a/deps/fmt
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit f5e54359df4c26b6230fc61d38aa294581393084
diff --git a/deps/json b/deps/json
deleted file mode 160000
index 4f8fba1406..0000000000
--- a/deps/json
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit 4f8fba14066156b73f1189a2b8bd568bde5284c5
diff --git a/deps/nccl b/deps/nccl
deleted file mode 160000
index 3996562690..0000000000
--- a/deps/nccl
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit 399656269027c1818fc999ccf8ec4dd838cec50d
diff --git a/deps/rapidcheck b/deps/rapidcheck
deleted file mode 160000
index 1505cbbce7..0000000000
--- a/deps/rapidcheck
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit 1505cbbce733bde3b78042cf2e9309c0b7f227a2
diff --git a/deps/spdlog b/deps/spdlog
deleted file mode 160000
index ad0e89cbfb..0000000000
--- a/deps/spdlog
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit ad0e89cbfb4d0c1ce4d097e134eb7be67baebb36
diff --git a/deps/visit_struct b/deps/visit_struct
deleted file mode 160000
index add1752b7a..0000000000
--- a/deps/visit_struct
+++ /dev/null
@@ -1 +0,0 @@
-Subproject commit add1752b7a1d806da4cee6aa135518003afd5bf7
diff --git a/flake.nix b/flake.nix
index da162eba26..3c06347214 100644
--- a/flake.nix
+++ b/flake.nix
@@ -81,17 +81,7 @@
       devShells = rec {
         ci = mkShell {
           shellHook = ''
-            export PATH="$HOME/ff/.scripts/:$PATH"
             export RC_PARAMS="max_discard_ratio=100"
-            export CMAKE_FLAGS="-DFF_USE_EXTERNAL_NCCL=ON \
-                                -DFF_USE_EXTERNAL_JSON=ON \
-                                -DFF_USE_EXTERNAL_FMT=ON \
-                                -DFF_USE_EXTERNAL_SPDLOG=ON \
-                                -DFF_USE_EXTERNAL_DOCTEST=ON \
-                                -DFF_USE_EXTERNAL_RAPIDCHECK=ON \
-                                -DFF_USE_EXTERNAL_EXPECTED=ON \
-                                -DFF_USE_EXTERNAL_GBENCHMARK=ON \
-                                -DFF_USE_EXTERNAL_LIBASSERT=ON"
           '';
 
           buildInputs = builtins.concatLists [
diff --git a/lib/CMakeLists.txt b/lib/CMakeLists.txt
index 501de709c1..c50e35620a 100644
--- a/lib/CMakeLists.txt
+++ b/lib/CMakeLists.txt
@@ -6,7 +6,7 @@ add_subdirectory(local-execution)
 add_subdirectory(realm-execution)
 add_subdirectory(task-spec)
 add_subdirectory(utils)
-add_subdirectory(ffi)
+#add_subdirectory(ffi)
 add_subdirectory(substitutions)
 add_subdirectory(substitution-generator)
 add_subdirectory(models)
diff --git a/lib/compiler/CMakeLists.txt b/lib/compiler/CMakeLists.txt
index 26fdc8a38a..61a8a16d79 100644
--- a/lib/compiler/CMakeLists.txt
+++ b/lib/compiler/CMakeLists.txt
@@ -10,9 +10,9 @@ ff_add_library(
   DEPS
     op-attrs
     utils
-    json
+    deps::json
     pcg
-    spdlog
+    deps::spdlog
     substitutions
 )
 
diff --git a/lib/compiler/test/CMakeLists.txt b/lib/compiler/test/CMakeLists.txt
index 3399a45f0f..9e2b532806 100644
--- a/lib/compiler/test/CMakeLists.txt
+++ b/lib/compiler/test/CMakeLists.txt
@@ -8,7 +8,7 @@ ff_add_test_executable(
   DEPS
     utils
     compiler
-    doctest
+    deps::doctest
     utils-test-common
     models
 )
diff --git a/lib/kernels/CMakeLists.txt b/lib/kernels/CMakeLists.txt
index f5d88f102f..d28cfbb94e 100644
--- a/lib/kernels/CMakeLists.txt
+++ b/lib/kernels/CMakeLists.txt
@@ -1,6 +1,6 @@
 set(project_target kernels)
 
-project(${project_target} 
+project(${project_target}
 	LANGUAGES CXX CUDA)
 
 file(GLOB_RECURSE SRC
@@ -25,9 +25,9 @@ target_include_directories(
 target_link_libraries(
   ${project_target}
   op-attrs
-  cuda
-  cudnn
-  nccl
+  deps::cublas
+  deps::cudnn
+  deps::nccl
   utils
   pcg
 )
@@ -35,8 +35,8 @@ target_link_libraries(
 define_ff_vars(${project_target})
 
 set_target_properties(
-  ${project_target} 
-  PROPERTIES 
+  ${project_target}
+  PROPERTIES
   CUDA_STANDARD 17
 )
 
diff --git a/lib/kernels/src/cuda/ops/attention_kernels.cu b/lib/kernels/src/cuda/ops/attention_kernels.cu
index a5fd9ea648..c38f26e002 100644
--- a/lib/kernels/src/cuda/ops/attention_kernels.cu
+++ b/lib/kernels/src/cuda/ops/attention_kernels.cu
@@ -55,7 +55,11 @@ MHAPerDeviceState gpu_init_kernel(PerDeviceFFHandle const &handle,
 
   // Currently do not support adding bias to key/value projection
   assert(!add_bias_kv);
+#if CUDNN_MAJOR >= 9
+  unsigned attnMode = CUDNN_ATTN_QUERYMAP_ALL_TO_ONE;
+#else
   cudnnAttnQueryMap_t attnMode = CUDNN_ATTN_QUERYMAP_ALL_TO_ONE;
+#endif
 
   // Assume no beam search for now
   int maxBeamSize = 1;
diff --git a/lib/kernels/test/CMakeLists.txt b/lib/kernels/test/CMakeLists.txt
index 981f87b3d8..c3ba1c34b8 100644
--- a/lib/kernels/test/CMakeLists.txt
+++ b/lib/kernels/test/CMakeLists.txt
@@ -3,16 +3,12 @@ ff_add_test_executable(
     kernels-tests
   SRC_PATTERNS
     src/*.cc
-  PRIVATE_INCLUDE 
+  PRIVATE_INCLUDE
     src/
   DEPS
-    doctest
+    deps::doctest
     utils-test-common
     kernels
     op-attrs
-    cuda
-    cudnn
-    cudart
-    cublas
     pcg
 )
diff --git a/lib/local-execution/CMakeLists.txt b/lib/local-execution/CMakeLists.txt
index b75f81fb3e..1f63838d76 100644
--- a/lib/local-execution/CMakeLists.txt
+++ b/lib/local-execution/CMakeLists.txt
@@ -13,7 +13,7 @@ ff_add_library(
     kernels
     task-spec
     pcg
-    spdlog
+    deps::spdlog
     compiler
 )
 
diff --git a/lib/local-execution/test/CMakeLists.txt b/lib/local-execution/test/CMakeLists.txt
index 0e79376575..e364ed3aee 100644
--- a/lib/local-execution/test/CMakeLists.txt
+++ b/lib/local-execution/test/CMakeLists.txt
@@ -3,10 +3,10 @@ ff_add_test_executable(
     local-execution-tests
   SRC_PATTERNS
     src/*.cc
-  PRIVATE_INCLUDE 
+  PRIVATE_INCLUDE
     src/
   DEPS
-    doctest
+    deps::doctest
     utils-test-common
     local-execution
     kernels
diff --git a/lib/models/CMakeLists.txt b/lib/models/CMakeLists.txt
index 4f4b22ed47..16e076886f 100644
--- a/lib/models/CMakeLists.txt
+++ b/lib/models/CMakeLists.txt
@@ -11,7 +11,7 @@ ff_add_library(
     op-attrs
     utils
     pcg
-    rapidcheck
+    deps::rapidcheck
 )
 
 add_subdirectory(test)
diff --git a/lib/models/test/CMakeLists.txt b/lib/models/test/CMakeLists.txt
index 9c2ae64d4e..349be009c8 100644
--- a/lib/models/test/CMakeLists.txt
+++ b/lib/models/test/CMakeLists.txt
@@ -7,6 +7,6 @@ ff_add_test_executable(
     src/
   DEPS
     models
-    doctest
+    deps::doctest
     utils-test-common
 )
diff --git a/lib/op-attrs/test/CMakeLists.txt b/lib/op-attrs/test/CMakeLists.txt
index b6ff72fc00..e79fdb45c8 100644
--- a/lib/op-attrs/test/CMakeLists.txt
+++ b/lib/op-attrs/test/CMakeLists.txt
@@ -8,6 +8,6 @@ ff_add_test_executable(
   DEPS
     utils
     op-attrs
-    doctest
+    deps::doctest
     utils-test-common
 )
diff --git a/lib/pcg/CMakeLists.txt b/lib/pcg/CMakeLists.txt
index e6eb182740..7ad518bc79 100644
--- a/lib/pcg/CMakeLists.txt
+++ b/lib/pcg/CMakeLists.txt
@@ -10,7 +10,7 @@ ff_add_library(
   DEPS
     op-attrs
     utils
-    rapidcheck
+    deps::rapidcheck
 )
 
 add_subdirectory(ffi)
diff --git a/lib/pcg/ffi/src/pcg.cc b/lib/pcg/ffi/src/pcg.cc
index 4e1bd55bc3..c8be9b47e7 100644
--- a/lib/pcg/ffi/src/pcg.cc
+++ b/lib/pcg/ffi/src/pcg.cc
@@ -1,2 +1,2 @@
 #include "flexflow/pcg.h"
-#include "pcg/model_compilation.h"
+// #include "pcg/model_compilation.h"
diff --git a/lib/pcg/test/CMakeLists.txt b/lib/pcg/test/CMakeLists.txt
index 685d1d8b88..3aae2ed6f6 100644
--- a/lib/pcg/test/CMakeLists.txt
+++ b/lib/pcg/test/CMakeLists.txt
@@ -8,6 +8,6 @@ ff_add_test_executable(
   DEPS
     utils
     pcg
-    doctest
+    deps::doctest
     utils-test-common
 )
diff --git a/lib/realm-execution/CMakeLists.txt b/lib/realm-execution/CMakeLists.txt
index 25a51ada54..49fbcfa4e0 100644
--- a/lib/realm-execution/CMakeLists.txt
+++ b/lib/realm-execution/CMakeLists.txt
@@ -13,10 +13,10 @@ ff_add_library(
     local-execution
     op-attrs
     pcg
-    spdlog
+    deps::spdlog
     task-spec
     utils
-    realm
+    deps::realm
 )
 
 add_subdirectory(test)
diff --git a/lib/realm-execution/test/CMakeLists.txt b/lib/realm-execution/test/CMakeLists.txt
index b3beff42c0..5cdd6c8282 100644
--- a/lib/realm-execution/test/CMakeLists.txt
+++ b/lib/realm-execution/test/CMakeLists.txt
@@ -6,7 +6,7 @@ ff_add_test_executable(
   PRIVATE_INCLUDE
     src/
   DEPS
-    doctest
+    deps::doctest
     utils-test-common
     realm-execution
     kernels
diff --git a/lib/substitution-generator/test/CMakeLists.txt b/lib/substitution-generator/test/CMakeLists.txt
index 166c7ab51f..0a61d2e049 100644
--- a/lib/substitution-generator/test/CMakeLists.txt
+++ b/lib/substitution-generator/test/CMakeLists.txt
@@ -13,7 +13,7 @@ ff_add_test_executable(
     src/
   DEPS
     utils
-    doctest
+    deps::doctest
     substitution-generator
     utils-test-common
 )
diff --git a/lib/substitutions/test/CMakeLists.txt b/lib/substitutions/test/CMakeLists.txt
index cfd6383e95..d11c2eab1a 100644
--- a/lib/substitutions/test/CMakeLists.txt
+++ b/lib/substitutions/test/CMakeLists.txt
@@ -7,7 +7,7 @@ ff_add_test_executable(
     src/
   DEPS
     utils
-    doctest
+    deps::doctest
     substitutions
     utils-test-common
 )
diff --git a/lib/task-spec/CMakeLists.txt b/lib/task-spec/CMakeLists.txt
index 3c7c91af67..af8f7f32e0 100644
--- a/lib/task-spec/CMakeLists.txt
+++ b/lib/task-spec/CMakeLists.txt
@@ -12,7 +12,7 @@ ff_add_library(
     utils
     kernels
     pcg
-    spdlog
+    deps::spdlog
     compiler
 )
 
diff --git a/lib/task-spec/test/CMakeLists.txt b/lib/task-spec/test/CMakeLists.txt
index 9665dba88e..36de69734e 100644
--- a/lib/task-spec/test/CMakeLists.txt
+++ b/lib/task-spec/test/CMakeLists.txt
@@ -6,9 +6,8 @@ ff_add_test_executable(
   PRIVATE_INCLUDE
     src/
   DEPS
-    doctest
+    deps::doctest
     utils-test-common
-    # local-execution
     kernels
     task-spec
     op-attrs
diff --git a/lib/utils/CMakeLists.txt b/lib/utils/CMakeLists.txt
index e2f7c433d6..73ac005ef2 100644
--- a/lib/utils/CMakeLists.txt
+++ b/lib/utils/CMakeLists.txt
@@ -8,13 +8,13 @@ ff_add_library(
   PRIVATE_INCLUDE
     src/
   DEPS
-    expected
-    fmt
-    json
-    cuda
-    rapidcheck
-    libassert
-)      
+    deps::expected
+    deps::fmt
+    deps::json
+    CUDA::cudart
+    deps::rapidcheck
+    deps::libassert
+)
 
 add_subdirectory(ffi)
 add_subdirectory(test)
diff --git a/lib/utils/test/CMakeLists.txt b/lib/utils/test/CMakeLists.txt
index 660ef99e87..fb2c7e0e07 100644
--- a/lib/utils/test/CMakeLists.txt
+++ b/lib/utils/test/CMakeLists.txt
@@ -7,7 +7,7 @@ ff_add_test_executable(
     src/
   DEPS
     utils
-    doctest
+    deps::doctest
     utils-test-common
 )
 
diff --git a/lib/utils/test/common/CMakeLists.txt b/lib/utils/test/common/CMakeLists.txt
index 18681af9c6..1e7abcaba3 100644
--- a/lib/utils/test/common/CMakeLists.txt
+++ b/lib/utils/test/common/CMakeLists.txt
@@ -9,6 +9,6 @@ ff_add_library(
     src/
   DEPS
     utils
-    rapidcheck
-    doctest
+    deps::rapidcheck
+    deps::doctest
 )