Update cpp lib to 1.3.1

.github/workflows/ci_requirements.txt

@@ -1,4 +1,4 @@
-# Copy of executorch/requirements-dev.txt as of v1.2.0
+# Copy of executorch/requirements-dev.txt as of v1.3.1
 
 cmake>=3.29, <4.0.0  # For building binary targets in the wheel.
 packaging>=24.2 # Lower bound required by setuptools
@@ -10,4 +10,4 @@ zstd  # Imported by resolve_buck.py.
 certifi  # Imported by resolve_buck.py.
 lintrunner==0.12.7
 lintrunner-adapters==0.13.0
-torch==2.11.0
+torch==2.12.0

.ruff.toml

@@ -0,0 +1,4 @@
+line-length = 120
+
+[lint]
+ignore = ["C408", "TRY002"]

README.md

@@ -22,6 +22,7 @@ import torch
 from torch.export import export
 from executorch.exir import to_edge_transform_and_lower
 
+
 class Add(torch.nn.Module):
     def __init__(self):
         super(Add, self).__init__()
@@ -62,7 +63,7 @@ See `example/hello_world` for a complete example.
 ## Build
 To use the library you must compile the C++ executorch library yourself, as there are many configurations that
 determines which modules, backends, and operations are supported. See the `executorch-sys` crate for more info.
-Currently the supported Cpp executorch version is `1.2.0`.
+Currently the supported Cpp executorch version is `1.3.1`.
 
 
 ## Cargo Features

etc/download_headers.py

@@ -37,6 +37,7 @@ def main():
             "runtime/core/**/*.h",
             "runtime/executor/**/*.h",
             "runtime/platform/**/*.h",
+            "runtime/kernel/operator_registry.h",
             "runtime/backend/options.h",
             "runtime/backend/backend_options_map.h",
             "extension/data_loader/**/*.h",
@@ -62,6 +63,8 @@ def main():
             "extension/data_loader/mman.h",
             "extension/data_loader/mman_windows.h",
             "extension/module/bundled_module.h",  # TODO
+            "runtime/core/device_allocator.h",  # TODO
+            "runtime/core/device_memory_buffer.h",  # TODO
             "extension/flat_tensor/serialize/serialize.h",
             "**/test/**",
             "**/testing_util/**",

etc/setup_dev_env.py

@@ -22,13 +22,12 @@ def main():
     parser.add_argument(
         "--skip-executorch-python",
         action="store_true",
-        help="Remove the existing executorch directory before cloning",
+        help="Skip installing the executorch Python package",
     )
     args = parser.parse_args()
 
-    if args.clean:
-        if DEV_EXECUTORCH_DIR.exists():
-            shutil.rmtree(DEV_EXECUTORCH_DIR)
+    if args.clean and DEV_EXECUTORCH_DIR.exists():
+        shutil.rmtree(DEV_EXECUTORCH_DIR)
 
     # TODO setup a venv here
 
@@ -50,7 +49,7 @@ def main():
                 "install",
                 "-r",
                 DEV_EXECUTORCH_DIR / "requirements-dev.txt",
-                "torch==2.11.0",
+                "torch==2.12.0",
                 "--extra-index-url",
                 "https://download.pytorch.org/whl/test/cpu",
             ]
@@ -72,7 +71,7 @@ def clone_executorch():
                 "--depth",
                 "1",
                 "--branch",
-                "v1.2.0",
+                "v1.3.1",
                 "https://github.com/pytorch/executorch.git",
                 ".",
             ],

examples/data_map/export_model.py

@@ -2,9 +2,8 @@
 from pathlib import Path
 
 import torch
-from torch.export import export
-
 from executorch.exir import ExecutorchBackendConfig, to_edge_transform_and_lower
+from torch.export import export
 
 
 class ModuleAddMul(torch.nn.Module):

examples/data_map/src/main.rs

@@ -73,7 +73,7 @@ fn main_program() {
     let data_map_loader = FileDataLoader::from_path(&data_file, None).unwrap();
     let data_map = FlatTensorDataMap::load(&data_map_loader).unwrap();
     let mut method = program
-        .load_method(c"forward", &memory_manager, None, Some(&data_map))
+        .load_method(c"forward", &memory_manager, None, Some(&data_map), None)
         .unwrap();
 
     let data = array![[1.0_f32, 2.0], [3.0, 4.0]];

examples/etdump/export_model.py

@@ -1,15 +1,14 @@
 from pathlib import Path
 
 import torch
-from torch.export import export
-
 from executorch.exir import to_edge_transform_and_lower
+from torch.export import export
 
 
 # A simple PyTorch model that adds two input tensors
 class Add(torch.nn.Module):
     def __init__(self):
-        super(Add, self).__init__()
+        super().__init__()
 
     def forward(self, x: torch.Tensor, y: torch.Tensor):
         return x + y

examples/hello_world/export_model.py

@@ -1,15 +1,14 @@
 from pathlib import Path
 
 import torch
-from torch.export import export
-
 from executorch.exir import to_edge_transform_and_lower
+from torch.export import export
 
 
 # A simple PyTorch model that adds two input tensors
 class Add(torch.nn.Module):
     def __init__(self):
-        super(Add, self).__init__()
+        super().__init__()
 
     def forward(self, x: torch.Tensor, y: torch.Tensor):
         return x + y

examples/llama3/README.md

@@ -72,4 +72,4 @@ To run the example, follow these steps (note that some steps should run from the
 The llama model can be exported with many options, such as quantization, different data types (f32, bf16), different backends, kv caching, etc.
 This example use a specific set of options, as specified above.
 Different options require different export commands and modifications to the code and build script which you can play around with.
-See the [llama README](https://github.com/pytorch/executorch/blob/v1.2.0/examples/models/llama/README.md) at the Cpp executorch repository for more details.
+See the [llama README](https://github.com/pytorch/executorch/blob/v1.3.1/examples/models/llama/README.md) at the Cpp executorch repository for more details.

examples/nano-gpt/model.py

@@ -7,14 +7,15 @@
 https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt2/modeling_gpt2.py
 """
 
-import math
 import inspect
+import math
 from dataclasses import dataclass
 
 import torch
-import torch.nn as nn
+from torch import nn
 from torch.nn import functional as F
 
+
 class LayerNorm(nn.Module):
     """ LayerNorm but with an optional bias. PyTorch doesn't support simply bias=False """
 
@@ -145,7 +146,7 @@ def __init__(self, config):
                 torch.nn.init.normal_(p, mean=0.0, std=0.02/math.sqrt(2 * config.n_layer))
 
         # report number of parameters
-        print("number of parameters: %.2fM" % (self.get_num_params()/1e6,))
+        print(f"number of parameters: {self.get_num_params()/1e6:.2f}M")
 
     def get_num_params(self, non_embedding=True):
         """
@@ -169,7 +170,7 @@ def _init_weights(self, module):
 
     def forward(self, idx, targets=None):
         device = idx.device
-        b, t = idx.size()
+        _b, t = idx.size()
         assert t <= self.config.block_size, f"Cannot forward sequence of length {t}, block size is only {self.config.block_size}"
         pos = torch.arange(0, t, dtype=torch.long, device=device) # shape (t)
 
@@ -210,7 +211,7 @@ def from_pretrained(cls, model_type, override_args=None):
         # only dropout can be overridden see more notes below
         assert all(k == 'dropout' for k in override_args)
         from transformers import GPT2LMHeadModel
-        print("loading weights from pretrained gpt: %s" % model_type)
+        print(f"loading weights from pretrained gpt: {model_type}")
 
         # n_layer, n_head and n_embd are determined from model_type
         config_args = {
@@ -280,7 +281,7 @@ def configure_optimizers(self, weight_decay, learning_rate, betas, device_type):
         # Create AdamW optimizer and use the fused version if it is available
         fused_available = 'fused' in inspect.signature(torch.optim.AdamW).parameters
         use_fused = fused_available and device_type == 'cuda'
-        extra_args = dict(fused=True) if use_fused else dict()
+        extra_args = {'fused': True} if use_fused else {}
         optimizer = torch.optim.AdamW(optim_groups, lr=learning_rate, betas=betas, **extra_args)
         print(f"using fused AdamW: {use_fused}")
 

examples/no_ndarray/export_model.py

@@ -1,15 +1,14 @@
 from pathlib import Path
 
 import torch
-from torch.export import export
-
 from executorch.exir import to_edge_transform_and_lower
+from torch.export import export
 
 
 # A simple PyTorch model that adds two input tensors
 class Add(torch.nn.Module):
     def __init__(self):
-        super(Add, self).__init__()
+        super().__init__()
 
     def forward(self, x: torch.Tensor, y: torch.Tensor):
         return x + y

examples/no_std/export_model.py

@@ -1,15 +1,14 @@
 from pathlib import Path
 
 import torch
-from torch.export import export
-
 from executorch.exir import to_edge_transform_and_lower
+from torch.export import export
 
 
 # A simple PyTorch model that adds two input tensors
 class Add(torch.nn.Module):
     def __init__(self):
-        super(Add, self).__init__()
+        super().__init__()
 
     def forward(self, x: torch.Tensor, y: torch.Tensor):
         return x + y

examples/no_std/src/main.rs

@@ -51,7 +51,7 @@ fn real_main() {
     let memory_manager = MemoryManager::new(&allocator, Some(&mut planned_memory), None);
 
     let mut method = program
-        .load_method(c"forward", &memory_manager, None, None)
+        .load_method(c"forward", &memory_manager, None, None, None)
         .unwrap();
 
     let input_array1 = ArrayStorage::new(array!(1.0_f32)).unwrap();

examples/raw_tensor/export_model.py

@@ -1,15 +1,14 @@
 from pathlib import Path
 
 import torch
-from torch.export import export
-
 from executorch.exir import to_edge_transform_and_lower
+from torch.export import export
 
 
 # A simple PyTorch model that adds two input tensors
 class Add(torch.nn.Module):
     def __init__(self):
-        super(Add, self).__init__()
+        super().__init__()
 
     def forward(self, x: torch.Tensor, y: torch.Tensor):
         return x + y

executorch-sys/README.md

@@ -4,14 +4,14 @@ For a general description of the project, see the the `executorch` crate.
 
 ## Build
 To build the library, you need to build the C++ library yourself first.
-Currently the supported Cpp executorch version is `1.2.0`.
+Currently the supported Cpp executorch version is `1.3.1`.
 The C++ library allow for great flexibility with many flags, customizing which modules, kernels, and extensions are built.
 Multiple static libraries are built, and the Rust library links to them.
 In the following example we build the C++ library with the necessary flags to run example `hello_world`:
 ```bash
 # Clone the C++ library
 cd ${EXECUTORCH_CPP_DIR}
-git clone --depth 1 --branch v1.2.0 https://github.com/pytorch/executorch.git .
+git clone --depth 1 --branch v1.3.1 https://github.com/pytorch/executorch.git .
 git submodule sync --recursive
 git submodule update --init --recursive
 

executorch-sys/build.rs

@@ -1,6 +1,6 @@
 use std::path::{Path, PathBuf};
 
-// const EXECUTORCH_VERSION: &str = "1.2.0";
+// const EXECUTORCH_VERSION: &str = "1.3.1";
 
 fn main() {
     // TODO: verify on runtime we use the correct version of executorch
@@ -98,6 +98,8 @@ fn generate_bindings() {
         .opaque_type("ET_MemoryManager")
         .opaque_type("ET_OptionalTensorStorage")
         .opaque_type("ET_DumpGen")
+        .opaque_type("ET_BackendOption")
+        .opaque_type("ET_LoadBackendOptionsMap")
         .blocklist_item("ET_FreeableBuffer")
         .blocklist_item(".*_bindgen_ty_.*")
         .parse_callbacks(Box::new(bindgen::CargoCallbacks::new()));